CN108615043A - A kind of video classification methods and system - Google Patents
A kind of video classification methods and system Download PDFInfo
- Publication number
- CN108615043A CN108615043A CN201611137674.1A CN201611137674A CN108615043A CN 108615043 A CN108615043 A CN 108615043A CN 201611137674 A CN201611137674 A CN 201611137674A CN 108615043 A CN108615043 A CN 108615043A
- Authority
- CN
- China
- Prior art keywords
- classification
- video
- frame
- key frame
- input video
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/758—Involving statistics of pixels or of feature values, e.g. histogram matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Studio Devices (AREA)
Abstract
The embodiment of the invention discloses a kind of video classification methods and system, the method includes:The system obtains at least one video segmentation from input video;The system obtains the corresponding key frame of each video segmentation according to the range performance in each video segmentation;The system carries out image classification to each key frame, obtains the corresponding static classification set of each key frame;The system obtains the classification results of the input video according to the corresponding key frame of each video segmentation, the corresponding static classification set of each key frame and default video classification parameter, and carrying out visual classification using key frame solves the problems, such as that visual classification accuracy is poor.
Description
Technical field
The present invention relates to technical field of video image processing more particularly to a kind of video classification methods and systems.
Background technology
With the explosive increase of video data, is handled for massive video data and extract having in video content
Imitating information becomes current research hotspot.Visual classification technology is one of the key technology of video content recognition and retrieval.
Low-level image feature and motion feature of the current visual classification technology based on video image, significantly regard for feature
Frequency can reach preferable classifying quality, but not satisfactory for the distant visual classification effect of feature.Therefore, having must
Video data is handled from video content, further increases the accuracy of classification.
Invention content
In order to solve the above technical problems, an embodiment of the present invention is intended to provide a kind of video classification methods and system, raising regards
The accuracy of frequency division class.
The technical proposal of the invention is realized in this way:
In a first aspect, an embodiment of the present invention provides a kind of video classification methods, the method is used for a kind of visual classification
System, the method includes:
The system obtains at least one video segmentation from input video;
The system obtains the corresponding key frame of each video segmentation according to the range performance in each video segmentation;
The system carries out image classification to each key frame, obtains the corresponding static classification set of each key frame;
The system is according to the corresponding key frame of each video segmentation, the corresponding static classification set of each key frame and default regards
Frequency classification parameter obtains the classification results of the input video.
In above-described embodiment, the system obtains at least one video segmentation from input video, specifically includes:
HS two-dimensional color Histogram distance and preset first threshold of the system according to adjacent image frame in input video
In correspondence and input video between value between the perceptual hash vector distance and preset second threshold of adjacent image frame
Correspondence input video is segmented, obtain at least one video segmentation.
Further, the system according to the HS two-dimensional colors Histogram distance of adjacent image frame in input video with it is default
First threshold between correspondence and input video in adjacent image frame perceptual hash vector distance and preset second
Correspondence between threshold value is segmented input video, obtains at least one video segmentation, specifically includes:
The tone saturation degree HS two-dimensional colors histogram of adjacent image frame and perception are breathed out in the system-computed input video
Uncommon vector;
The system is vectorial according to the HS two-dimensional colors histogram and perceptual hash of adjacent image frame in the input video,
Calculate the HS two-dimensional colors Histogram distance and perceptual hash vector distance of adjacent image frame in the input video;
When adjacent image frame in the input video HS two-dimensional color Histogram distances be more than preset first threshold, and
When the perceptual hash vector distance of adjacent image frame is more than preset second threshold in the input video, the system is to input
Video is segmented, and at least one video segmentation is obtained.
Further, the process of the system-computed HS two-dimensional color histograms specifically includes:
Picture frame is transformed into tone saturation degree lightness hsv color space by the system from RGB RGB color, is obtained
Take tone H components, saturation degree S components;
The corresponding channels H of H components channel S corresponding with S components is divided into a sections by the system, and statistics is schemed
As the HS two-dimensional color histograms of frame.
Further, the process of the system-computed perceptual hash vector specifically includes:
Picture frame is zoomed to pre-set dimension by the system, obtains the first image array;
The system carries out image gray processing processing to described first image matrix, obtains the second image array;
The system carries out two-dimension discrete cosine transform DCT to second image array, obtains third image array;
The system, which is chosen, presets submatrix as the 4th image array in third image array;
The average pixel value of 4th image array described in the system-computed;
The pixel value of each picture element in 4th image array is compared by the system with the average pixel value,
Obtain newer 4th image array;
Newer 4th image array is carried out vectorization by the system, obtains final perceptual hash vector.
In above-described embodiment, when the adjacent image frame HS two-dimensional colors Histogram distance and preset first threshold it
Between correspondence and adjacent image frame perceptual hash vector distance and preset second threshold between correspondence not
The HS two-dimensional color Histogram distances for meeting the adjacent image frame are more than preset first threshold, and the perception of adjacent image frame
When Hash vector distance is more than preset second threshold, the system will not be split input video.
In above-described embodiment, it is corresponding to obtain each video segmentation according to the range performance in each video segmentation for the system
Key frame specifically includes:
The system obtains the average HS two-dimensional colors histogram of each video segmentation, and straight according to each average HS two-dimensional colors
Between the HS two-dimensional color histograms of the picture frame of side's figure video segmentation corresponding with each average HS two-dimensional color histograms away from
From relationship, the corresponding key frame of each video segmentation is obtained.
Further, the system is by the average HS two-dimensional colors histogram of video segmentation and average HS two-dimensional color histograms
The HS two-dimensional color histograms for scheming picture frame all in corresponding video segmentation are compared, video segmentation described in selected distance
The average nearest picture frame of HS two-dimensional colors histogram key frame of the HS two-dimensional colors histogram as the video segmentation.
In above-described embodiment, the system carries out image classification to each key frame, and it is static point corresponding to obtain each key frame
Class set, specifically includes:
The system carries out image classification by Image Classifier to each key frame, is obtained according to default key frame classification parameter
Take the corresponding static classification set of each key frame;Wherein, described image grader is generated by deep neural network.
Further, the default key frame classification parameter, for limiting in the corresponding static classification set of each key frame
The number of crucial frame category.
In upper embodiment, the system is according to the corresponding key frame of each video segmentation, the corresponding static classification of each key frame
Set and default video classification parameter, obtain the classification results of the input video, specifically include:
The number of image frames calculating that the number of image frames and each video segmentation that the system includes according to input video include respectively regards
The time weighting of frequency division section;
The system obtains the visual classification of the input video according to the corresponding static classification set of each key frame
Set;
The system is according to each visual classification static classification set corresponding with each key frame in the visual classification set
Between relationship, obtain the visual classification coefficient of each visual classification in the visual classification set;
The system is regarded according to each visual classification in the time weighting of each video segmentation, the visual classification set
Frequency classification factor calculates the visual classification weight of each visual classification in visual classification set;
The system is according to the visual classification weight of each visual classification and default video class in the visual classification set
Other parameter obtains the final classification result of input video.
Further, the system is using the corresponding static classification union of sets collection of each key frame as the input video
Visual classification set.
Further, the default video classification parameter, the number for limiting input video classification results.
Second aspect, an embodiment of the present invention provides a kind of video classification system, the system comprises:First obtains mould
Block, the second acquisition module, the first sort module and the second sort module, wherein
First acquisition module, for obtaining at least one video segmentation from input video;
Second acquisition module, for according to the range performance in each video segmentation, it is corresponding to obtain each video segmentation
Key frame;
It is static point corresponding to obtain each key frame for carrying out image classification to each key frame for first sort module
Class set;
Second sort module, for static point corresponding according to the corresponding key frame of each video segmentation, each key frame
Class set and default video classification parameter, obtain the classification results of the input video.
In above-described embodiment, first acquisition module is specifically used for
According between the HS two-dimensional colors Histogram distance and preset first threshold of adjacent image frame in input video
Corresponding pass in correspondence and input video between the perceptual hash vector distance of adjacent image frame and preset second threshold
System is segmented input video, obtains at least one video segmentation.
Further, first acquisition module, is specifically used for
Calculate the HS two-dimensional colors histogram of adjacent image frame and perceptual hash vector in input video;
It is vectorial according to the HS two-dimensional colors histogram of adjacent image frame in the input video and perceptual hash, described in calculating
The HS two-dimensional colors Histogram distance and perceptual hash vector distance of adjacent image frame in input video;
When adjacent image frame in the input video HS two-dimensional color Histogram distances be more than preset first threshold, and
When the perceptual hash vector distance of adjacent image frame is more than preset second threshold in the input video, input video is carried out
Segmentation, obtains at least one video segmentation.
Further, for the process of calculating HS two-dimensional color histograms, first acquisition module includes the first transformation
Submodule and statistic submodule, wherein
First transformation submodule obtains H points for picture frame to be transformed into hsv color space from RGB color
Amount, S components;
The statistic submodule, for the corresponding channels H of H components channel S corresponding with S components to be divided into a sections,
Statistics obtains the HS two-dimensional color histograms of picture frame.
Further, for calculating the process of perceptual hash vector, the first acquisition module described in first acquisition module
Further include scaling submodule, gray processing processing submodule, the second transformation submodule, to choose submodule, computational submodule, comparison sub
Module and vectorization submodule, wherein
The scaling submodule obtains the first image array for picture frame to be zoomed to pre-set dimension;
The gray processing handles submodule, for carrying out image gray processing processing to described first image matrix, obtains the
Two image arrays;
Second transformation submodule obtains third image moment for carrying out two-dimensional dct to second image array
Battle array;
The selection submodule presets submatrix as the 4th image array for choosing in third image array;
The computational submodule, the average pixel value for calculating the 4th image array;
The comparison sub-module is used for the pixel value of each picture element in the 4th image array and the average pixel
Value is compared, and obtains newer 4th image array;
The vectorization submodule obtains final sense for newer 4th image array to be carried out vectorization
Know Hash vector.
Further, when between the HS two-dimensional colors Histogram distance and preset first threshold of the adjacent image frame
Correspondence between correspondence and the perceptual hash vector distance and preset second threshold of adjacent image frame is unsatisfactory for
The HS two-dimensional color Histogram distances of the adjacent image frame are more than preset first threshold, and the perceptual hash of adjacent image frame
When vector distance is more than preset second threshold, first acquisition module will not be split input video.
In above-described embodiment, second acquisition module is specifically used for
The average HS two-dimensional colors histogram of each video segmentation is obtained, and according to each average HS two-dimensional colors histogram and respectively
Averagely the distance between HS two-dimensional color histograms of picture frame of the corresponding video segmentation of HS two-dimensional color histograms relationship, is obtained
Take the corresponding key frame of each video segmentation.
Further, second acquisition module, is specifically used for
By the average HS two-dimensional colors histogram of video segmentation video segmentation corresponding with average HS two-dimensional color histograms
In the HS two-dimensional color histograms of all picture frames be compared, the average HS two-dimensional colors of video segmentation described in selected distance
Key frame of the HS two-dimensional colors histogram of the nearest picture frame of histogram as the video segmentation.
In above-described embodiment, first sort module is specifically used for
Image classification is carried out to each key frame by Image Classifier, each key is obtained according to default key frame classification parameter
The corresponding static classification set of frame;Wherein, described image grader is generated by deep neural network.
Further, the default key frame classification parameter, for limiting in the corresponding static classification set of each key frame
The number of crucial frame category.
In above-described embodiment, second sort module is obtained including the first acquisition submodule, the second acquisition submodule, third
Take submodule, the 4th acquisition submodule and the 5th acquisition submodule, wherein
First acquisition submodule, the figure that number of image frames and each video segmentation for including according to input video include
As frame number calculates the time weighting of each video segmentation;
Second acquisition submodule, for according to the corresponding static classification set of each key frame, obtaining described defeated
Enter the visual classification set of video;
The third acquisition submodule, for corresponding with each key frame according to each visual classification in the visual classification set
Static classification set between relationship, obtain the visual classification coefficient of each visual classification in the visual classification set;
4th acquisition submodule, for the time weighting according to each video segmentation, the visual classification set
In each visual classification visual classification coefficient, calculate visual classification set in each visual classification visual classification weight;
5th acquisition submodule, for the visual classification weight according to each visual classification in the visual classification set
And default video classification parameter, obtain the final classification result of input video.
Further, second acquisition submodule, is specifically used for
Using the corresponding static classification union of sets collection of each key frame as the visual classification set of the input video.
Further, in the 5th acquisition submodule, the default video classification parameter, for limiting input video point
The number of class result.
An embodiment of the present invention provides a kind of video classification methods and system, the method includes:The system is from input
At least one video segmentation is obtained in video;The system obtains each video segmentation according to the range performance in each video segmentation
Corresponding key frame;The system carries out image classification to each key frame, obtains the corresponding static classification set of each key frame;Institute
System is stated according to the corresponding key frame of each video segmentation, the corresponding static classification set of each key frame and default video classification to be joined
Number, obtains the classification results of the input video, and carrying out visual classification using key frame solves asking for visual classification accuracy difference
Topic.
Description of the drawings
Fig. 1 is a kind of flow chart of video classification methods provided in an embodiment of the present invention;
Fig. 2 is a kind of flow chart obtaining video segmentation provided in an embodiment of the present invention;
Fig. 3 is a kind of flow chart calculating HS two-dimensional color histograms provided in an embodiment of the present invention;
Fig. 4 is a kind of flow chart calculating perceptual hash vector provided in an embodiment of the present invention;
Fig. 5 is a kind of flow chart obtaining input video classification results provided in an embodiment of the present invention;
Fig. 6 is a kind of flow chart of visual classification specific method provided in an embodiment of the present invention;
Fig. 7 is a kind of particular flow sheet calculating HS two-dimensional color histograms provided in an embodiment of the present invention;
Fig. 8 is a kind of particular flow sheet calculating perceptual hash vector provided in an embodiment of the present invention.
Fig. 9 is a kind of structure diagram of video classification system provided in an embodiment of the present invention;
Figure 10 is a kind of structure diagram of first acquisition module provided in an embodiment of the present invention;
Figure 11 is the structure diagram of another first acquisition module provided in an embodiment of the present invention;
Figure 12 is a kind of structure diagram of second sort module provided in an embodiment of the present invention.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes.
The basic thought of the embodiment of the present invention is:The system obtains at least one video segmentation from input video;Institute
System is stated according to the range performance in each video segmentation, obtains the corresponding key frame of each video segmentation;The system is to each key
Frame carries out image classification, obtains the corresponding static classification set of each key frame;The system is according to the corresponding pass of each video segmentation
The corresponding static classification set of key frame, each key frame and default video classification parameter, obtain the classification results of the input video,
Visual classification is carried out using key frame and solves the problems, such as that visual classification accuracy is poor.
Embodiment one
Referring to Fig. 1, it illustrates a kind of video classification methods, the method is used for a kind of video classification system, the side
Method includes:
S101:The system obtains at least one video segmentation from input video;
S102:The system obtains the corresponding key frame of each video segmentation according to the range performance in each video segmentation;
S103:The system carries out image classification to each key frame, obtains the corresponding static classification set of each key frame;
S104:The system according to the corresponding key frame of each video segmentation, the corresponding static classification set of each key frame and
Default video classification parameter, obtains the classification results of the input video.
For step S101, the system obtains at least one video segmentation from input video, specifically includes:
HS two-dimensional color Histogram distance and preset first threshold of the system according to adjacent image frame in input video
In correspondence and input video between value between the perceptual hash vector distance and preset second threshold of adjacent image frame
Correspondence input video is segmented, obtain at least one video segmentation.
As shown in Fig. 2, the system according to the HS two-dimensional colors Histogram distance of adjacent image frame in input video with it is pre-
If first threshold between correspondence and input video in adjacent image frame perceptual hash vector distance and preset the
Correspondence between two threshold values is segmented input video, obtains at least one video segmentation, specifically includes:
S1011:In the system-computed input video HS two-dimensional colors histogram of adjacent image frame and perceptual hash to
Amount;
S1012:The system is breathed out according to the HS two-dimensional colors histogram and perception of adjacent image frame in the input video
Uncommon vector, calculates the HS two-dimensional colors Histogram distance and perceptual hash vector distance of adjacent image frame in the input video;
S1013:When the HS two-dimensional color Histogram distances of adjacent image frame in the input video are more than preset first
Threshold value, and in the input video adjacent image frame perceptual hash vector distance be more than preset second threshold when, the system
System is segmented input video, obtains at least one video segmentation.
For step S1011, by taking a picture frame in adjacent image frame in input video as an example, referring to Fig. 3, the system
Statistics calculate HS two-dimensional color histograms process be step S301 to step S302, specifically include:
S301:The system satisfies picture frame from RGB (RGB, Red Green Blue) color space conversion to tone
With degree lightness (HSV, Hue Saturation Value) color space, obtain tone (H, Hue) component, saturation degree (S,
Saturation) component;
S302:The corresponding channels H of H components channel S corresponding with S components is divided into a sections by the system, and statistics obtains
Obtain tone saturation degree (HS, Hue Saturation) two-dimensional color histogram of picture frame.
Specifically, it for step S301, is transformed into hsv color space from RGB color and obtains H components and S components
Formula is respectively as shown in formula (1) and formula (2);
In formula (1) and formula (2), R indicates that the red component of input video, G indicate the green component of input video,
B indicates that the blue component of input video, H indicate that the chrominance component of input video, S indicate the saturation degree component of input video, V tables
Show the lightness component of input video, V=max (R, G, B), 0≤V≤1;
Wherein, 0≤H≤360 in formula (1), if H<0, then H=H+360;In formula (2), 0≤S≤1.
For step S302, it is preferable that the system is by the corresponding channels H of H components channel S corresponding with S components
It is divided into 32 sections.
For step S1011, by taking a picture frame in adjacent image frame in input video as an example, referring to Fig. 4, the system
Statistics calculate perceptual hash vector process be step S401 to step S407, specifically include:
S401:Picture frame is zoomed to pre-set dimension by the system, obtains the first image array;
S402:The system carries out image gray processing processing to described first image matrix, obtains the second image array;
S403:The system carries out two-dimensional dct to second image array, obtains third image array;
S404:The system, which is chosen, presets submatrix as the 4th image array in third image array;
S405:The average pixel value of 4th image array described in the system-computed;
S406:The system carries out the pixel value of each picture element in the 4th image array and the average pixel value
Compare, obtains newer 4th image array;
S407:Newer 4th image array is carried out vectorization by the system, obtain final perceptual hash to
Amount.
For step S401, it is preferable that picture frame is zoomed to the size of 32 × 32cm by the system.
Specifically, for step S403, formula is such as discrete cosine transform (DCT, Discrete Cosine Transform)
Shown in formula (3);
Wherein, u, v=0,1,2 ..., N-1, it is the size value that picture frame is zoomed to pre-set dimension by the system, I to take N
(x, y) indicates image in the pixel value of (x, y) point, dct transform result of F (u, the v) expressions positioned at (u, v) point.
For step S404, it is preferable that the submatrix that the system chooses 8 × 8cm of the upper left corner in third image array is made
For the 4th image array.
Specifically, for step S406, when the pixel value of the picture element in the 4th image array is more than the average pixel
The pixel value of picture element in 4th image array is labeled as 0 by value, the system;Pixel in the 4th image array
The pixel value of point is less than the average pixel value, and the system marks the pixel value of the picture element in the 4th image array
It is 1, to obtain newer 4th image array.
For step S1012, shown in the calculating such as formula (4) of the HS two-dimensional colors Histogram distance;
Wherein, DHIndicate the two-dimensional color Histogram distance of adjacent image frame, Ht(x, y) indicates in adjacent image frame one
The HS two-dimensional colors histogram of picture frame is in the statistical value of (x, y) point, Ht+1(x, y) indicates another image in adjacent image frame
Statistical value of the HS two-dimensional colors histogram of frame in (x, y) point.
For step S1012, shown in the calculation formula such as formula (5) and formula (6) of the perceptual hash vector distance:
In formula (5), DpIndicate perceptual hash vector distance, Pt(i) perception of a picture frame in adjacent image frame is indicated
I-th of element of Hash vector, Pt+1(i) vectorial i-th of the element of the perceptual hash of another picture frame in adjacent image frame is indicated;
In formula (6), P is worked as in D (x, x') expressionst+1(i)=Pt+1(i) when, D (x, x')=1;Work as Pt+1(i)≠Pt+1(i) when, D (x, x')
=0.
For step S1013, it should be noted that when the adjacent image frame HS two-dimensional colors Histogram distance with it is pre-
If first threshold between correspondence and adjacent image frame perceptual hash vector distance and preset second threshold it
Between correspondence be unsatisfactory for the HS two-dimensional color Histogram distances of the adjacent image frame and be more than preset first threshold, and phase
When the perceptual hash vector distance of adjacent picture frame is more than preset second threshold, the system will not divide input video
It cuts.
For step S102, it is corresponding to obtain each video segmentation according to the range performance in each video segmentation for the system
Key frame specifically includes:
The system obtains the average HS two-dimensional colors histogram of each video segmentation, and straight according to each average HS two-dimensional colors
Between the HS two-dimensional color histograms of the picture frame of side's figure video segmentation corresponding with each average HS two-dimensional color histograms away from
From relationship, the corresponding key frame of each video segmentation is obtained.
Specifically, for step S102, the system is by the average HS two-dimensional colors histogram of video segmentation and average HS
The HS two-dimensional color histograms of all picture frames are compared in the corresponding video segmentation of two-dimensional color histogram, selected distance
The HS two-dimensional colors histogram of the average nearest picture frame of HS two-dimensional colors histogram of the video segmentation is as the video
The key frame of segmentation.
For step S103, the system carries out image classification to each key frame, and it is static point corresponding to obtain each key frame
Class set, specifically includes:
The system carries out image classification by Image Classifier to each key frame, is obtained according to default key frame classification parameter
Take the corresponding static classification set of each key frame;Wherein, described image grader is generated by deep neural network;
Specifically, for the generation of Image Classifier, the system collects each other data representing image of video class, makees
It, can also be by carrying out the pre- places such as image enhancement, rotation, random cropping to data representing image for the training data of disaggregated model
Training data is inputted deep neural network, the depth nerve by reason operation, further training for promotion data bulk, the system
The spy of the similar model extraction training data of depth convolutional neural networks such as AlexNet, GoogleNet or other can be used in network
Sign, by continuous repetitive exercise, obtains accurate image classification model as Image Classifier.
For step S103, the default key frame classification parameter, for limiting the corresponding static classification collection of each key frame
The number of crucial frame category in conjunction;
Wherein, the default key frame classification parameter needs are specifically determined according to actual conditions, for example, by successive ignition
Trained deep neural network, can be using the larger preceding k classes of probability in the classification results of deep neural network output as input
The static classification result of key frame of video.
For step S104, referring to Fig. 5, the system is corresponded to according to the corresponding key frame of each video segmentation, each key frame
Static classification set and default video classification parameter, obtain the classification results of the input video, specifically include:
S1041:The number of image frames meter that the number of image frames and each video segmentation that the system includes according to input video include
Calculate the time weighting of each video segmentation;
S1042:The system obtains regarding for the input video according to the corresponding static classification set of each key frame
Frequency division class set;
S1043:Static point corresponding with each key frame according to each visual classification in the visual classification set of the system
Relationship between class set obtains the visual classification coefficient of each visual classification in the visual classification set;
S1044:The system is according to each video point in the time weighting of each video segmentation, the visual classification set
The visual classification coefficient of class calculates the visual classification weight of each visual classification in visual classification set;
S1045:The system is according to the visual classification weight of each visual classification in the visual classification set and presets
Video classification parameter obtains the final classification result of input video.
For step S1041, it is assumed that input video is made of N number of picture frame, wherein key frame KF1,KF2,...,KFnIn
KFiCorresponding video segmentation SiBy niA picture frame composition, KFiCorresponding video segmentation SiTime weighting wiCalculation formula is such as public
Shown in formula (7).
For step S1042, specifically, the system is using the corresponding static classification union of sets collection of each key frame as institute
State the visual classification set of input video.
For step S1043, visual classification coefficient indicates key frame KFiStatic classification result in whether include video point
Visual classification C in class setx;As input video key KFiStatic classification result in include the video in visual classification set
Classify Cx, then Si(Cx)=1;As input video key frame KFiStatic classification result in do not include visual classification set in regarding
Frequency division class Cx, then Si(Cx)=0.
For step S1044, C in the visual classification set is calculatedxVisual classification weightAs shown in formula (8);
Wherein, wiFor KFiCorresponding video segmentation SiTime weighting, Si(Cx) indicate CxVisual classification coefficient.
For step S1045, the default video classification parameter, the number for limiting input video classification results;
Wherein, the default video classification parameter needs are specifically determined according to actual conditions, for example, can be by input video
The corresponding visual classification weight of visual classification set in the larger first k' final classification result as input video of weights.
A kind of video classification methods are present embodiments provided, the system obtains at least one video point from input video
Section;The system obtains the corresponding key frame of each video segmentation according to the range performance in each video segmentation;The system is to each
Key frame carries out image classification, obtains the corresponding static classification set of each key frame;The system is corresponded to according to each video segmentation
Key frame, the corresponding static classification set of each key frame and default video classification parameter, obtain the classification of the input video
As a result, carrying out visual classification using key frame solves the problems, such as that visual classification accuracy is poor.
Embodiment two
Based on the identical technical concept of previous embodiment, referring to Fig. 6, it illustrates a kind of visual classification specific method, institutes
The method of stating includes:
S601:In the system-computed input video HS two-dimensional colors histogram of adjacent image frame and perceptual hash to
Amount;
S602:The system is according to the HS two-dimensional colors histogram and perceptual hash of adjacent image frame in the input video
Vector calculates the HS two-dimensional colors Histogram distance and perceptual hash vector distance of adjacent image frame in the input video;
S603:When the HS two-dimensional color Histogram distances of adjacent image frame in the input video are more than preset first threshold
Value, and in the input video adjacent image frame perceptual hash vector distance be more than preset second threshold when, the system
Input video is segmented, at least one video segmentation is obtained;
S604:The system obtains the average HS two-dimensional colors histogram of each video segmentation, and according to each average HS two dimensions
The HS two-dimensional colors histogram of the picture frame of color histogram video segmentation corresponding with each average HS two-dimensional color histograms it
Between distance relation, obtain the corresponding key frame of each video segmentation;
S605:The system carries out image classification by Image Classifier to each key frame, according to default crucial frame category
The corresponding static classification set of each key frame of parameter acquiring;Wherein, described image grader is generated by deep neural network;
S606:The number of image frames that the number of image frames and each video segmentation that the system includes according to input video include calculates
The time weighting of each video segmentation;
S607:The system obtains regarding for the input video according to the corresponding static classification set of each key frame
Frequency division class set;
S608:The system is according to each visual classification static classification corresponding with each key frame in the visual classification set
Relationship between set obtains the visual classification coefficient of each visual classification in the visual classification set;
S609:The system is according to each video point in the time weighting of each video segmentation, the visual classification set
The visual classification coefficient of class calculates the visual classification weight of each visual classification in visual classification set;
S610:The system is according to the visual classification weight of each visual classification in the visual classification set and default regards
Frequency classification parameter obtains the final classification result of input video.
For step S601, by taking a picture frame in adjacent image frame in input video as an example, referring to Fig. 7, the system
Statistics calculate HS two-dimensional color histograms process be step S701 to step S702, specifically include:
S701:Picture frame is transformed into hsv color space by the system from RGB color, obtains H, component, S components;
S702:The corresponding channels H of H components channel S corresponding with S components is divided into a sections by the system, and statistics obtains
Obtain the HS two-dimensional color histograms of picture frame;Preferably, the system is corresponding with S components by the corresponding channels H of the H components
Channel S is divided into 32 sections.
Specifically, it for step S701, is transformed into hsv color space from RGB color and obtains H components and S components
Formula is respectively as shown in formula (1) and formula (2);
In formula (1) and formula (2), R indicates that the red component of input video, G indicate the green component of input video,
B indicates that the blue component of input video, H indicate that the chrominance component of input video, S indicate the saturation degree component of input video, V tables
Show the lightness component of input video, V=max (R, G, B), 0≤V≤1;Wherein, 0≤H≤360 in formula (1), if H<0, then H
=H+360;In formula (2), 0≤S≤1.
32 sections are divided into [0,360] section for step S702, such as by the channels H, specially [0,11.25),
[11.25,22.5) ... [348.75,360];Channel S is divided into 32 sections in [0,1] section, specially [0,0.03125),
[0.03125,0.0625) ... [0.96875,1];When the numerical value of H components belongs to [0,11.25), [11.25,22.5) ...
In [348.75,360] when some section, the statistical value in corresponding section adds 1, when the numerical value of S components belongs to [0,0.03125),
[0.03125,0.0625) ... in [0.96875,1] when some section, the statistical value in corresponding section adds 1, and final statistics obtains
Obtain the HS two-dimensional color histograms of picture frame.
For step S601, by taking a picture frame in adjacent image frame in input video as an example, referring to Fig. 8, sense is calculated
Know Hash vector process be step S801 to step S807, specifically include:
S801:Picture frame is zoomed to pre-set dimension by the system, obtains the first image array;Preferably, pre-set dimension
For 32 × 32cm;
S802:The system carries out image gray processing processing to described first image matrix, obtains the second image array;
S803:The system carries out two-dimensional dct transform to second image array, obtains third image array;
S804:The system, which is chosen, presets submatrix as the 4th image array in third image array;Preferably, it presets
Submatrix is the image array in the upper left corner 8 × 8 in third image array;
S805:The average pixel value of 4th image array described in the system-computed;
S806:The system carries out the pixel value of each picture element in the 4th image array and the average pixel value
Compare, obtains newer 4th image array;
S807:Newer 4th image array is carried out vectorization by the system, obtain final perceptual hash to
Amount.
Specifically, for step S803, shown in the formula such as formula (3) of DCT;
Wherein, u, v=0,1,2 ..., N-1, it is the size value that picture frame is zoomed to pre-set dimension by the system, I to take N
(x, y) indicates image in the pixel value of (x, y) point, DCT result of F (u, the v) expressions positioned at (u, v) point.
Specifically, for step S806, when the pixel value of the picture element in the 4th image array is more than the average pixel
The pixel value of picture element in 4th image array is labeled as 0 by value, the system;Pixel in the 4th image array
The pixel value of point is less than the average pixel value, and the system marks the pixel value of the picture element in the 4th image array
It is 1, to obtain newer 4th image array.
Specifically, for step S807, two values matrix of the system by newer 4th image array from 8 × 8
It is changed into 1 × 82Row matrix or be changed into 82× 1 column matrix, then 1 × 8 after changing2Row matrix or 82× 1 row square
Battle array is that final perceptual hash is vectorial.
For step S602, shown in the calculation formula such as formula (4) of the HS two-dimensional colors Histogram distance;
Wherein, DHIndicate the two-dimensional color Histogram distance of adjacent image frame, Ht(x, y) indicates in adjacent image frame one
The HS two-dimensional colors histogram of picture frame is in the statistical value of (x, y) point, Ht+1(x, y) indicates another image in adjacent image frame
Statistical value of the HS two-dimensional colors histogram of frame in (x, y) point;
Specifically, from formula (4) as can be seen that DHResult be two picture frames HS two-dimensional color histograms it is same
The quadratic sum of the difference of the statistical value of position (x, y) point.
For step S602, shown in the calculation formula such as formula (5) and formula (6) of the perceptual hash vector distance;
In formula (5), DpIndicate perceptual hash vector distance, Pt(i) perception of a picture frame in adjacent image frame is indicated
I-th of element of Hash vector, Pt+1(i) vectorial i-th of the element of the perceptual hash of another picture frame in adjacent image frame is indicated;
In formula (6), D (x, x') indicates Pt(i) and Pt+1(i) relationship between, works as Pt+1(i)=Pt+1(i) when, D (x, x')=1;When
Pt+1(i)≠Pt+1(i) when, D (x, x')=0;
Specifically, it can be seen that from formula (6) when vectorial i-th of the element of the perceptual hash of adjacent image frame is equal, D
(x, x')=1, when vectorial i-th of the element of the perceptual hash of adjacent image frame is unequal, D (x, x')=0, i.e., by adjacent image
The perceptual hash vector of frame carries out " position with ", then the result of " position with " is summed, obtained result is perceptual hash vector
Distance Dp。
For step S603, it should be noted that when the adjacent image frame HS two-dimensional colors Histogram distance with it is pre-
If first threshold between correspondence and adjacent image frame perceptual hash vector distance and preset second threshold it
Between correspondence be unsatisfactory for the HS two-dimensional color Histogram distances of the adjacent image frame and be more than preset first threshold, and phase
When the perceptual hash vector distance of adjacent picture frame is more than preset second threshold, the system will not divide input video
It cuts.
Specifically, for step S604, the system is by the average HS two-dimensional colors histogram of video segmentation and average HS
The HS two-dimensional color histograms of all picture frames are compared in the corresponding video segmentation of two-dimensional color histogram, selected distance
The HS two-dimensional colors histogram of the average nearest picture frame of HS two-dimensional colors histogram of the video segmentation is as the video
The key frame of segmentation.
Illustratively, the adjacent image frame F obtained for step S602 to step S604, the systemtWith Ft+1HS bis-
Tie up color histogram map distance DHMore than two-dimensional color histogram thresholding T1, and perceptual hash vector distance DpMore than perceptual hash to
Measure threshold value T2, then the system will be split input video, obtain a video segmentation, same method, final institute
State all video segmentation S that system will obtain the input video1,S2,...,Sn, the system-computed video segmentation SiIt is flat
Equal HS two-dimensional color histogramsAnd it willWith video segmentation SiIn the HS two-dimensional color histograms of all picture frames compared
Compared with, selection withHS two-dimensional colors histogram apart from immediate video segmentation picture frame is as video segmentation SiKey
Frame KFi, same method, the final system will obtain all video segmentation S1,S2,...,SnCorresponding key frame KF1,
KF2,...,KFn。
For step S605, the default key frame classification parameter, for limiting the corresponding static classification collection of each key frame
The number of crucial frame category in conjunction;
Wherein, the default key frame classification parameter needs are specifically determined according to actual conditions, for example, by successive ignition
Trained deep neural network, can be using the larger preceding k classes of probability in the classification results of deep neural network output as input
The static classification result of key frame of video.
For step S605, specifically, the generation for Image Classifier, the system collects each video class other generation
Table image data, as the training data of disaggregated model, can also by data representing image carry out image enhancement, rotation,
Training data is inputted depth nerve net by the pretreatment operations such as random cropping, further training for promotion data bulk, the system
The similar model of depth convolutional neural networks such as AlexNet, GoogleNet or other can be used in network, the deep neural network
The feature of extraction training data obtains accurate image classification model as Image Classifier by continuous repetitive exercise.
For step S606, it is assumed that input video is made of N number of picture frame, wherein key frame KF1,KF2,...,KFnIn
KFiCorresponding video segmentation SiBy niA picture frame composition, KFiCorresponding video segmentation SiTime weighting calculation formula such as formula
(7) shown in.
For step S607, specifically, the system is using the corresponding static classification union of sets collection of each key frame as institute
State the visual classification set of input video;
For example, default key frame classification parameter is 3, KF1Corresponding static classification collection is combined into { C1,C2,C5, KF2It is corresponding
Static classification collection is combined into { C2,C4,C6..., KFnCorresponding static classification collection is combined into { C3,C4,C5, then the video of input video
Category set is combined into { C1,C2,C3,C4,C5,C6}。
Specifically, for step S608, it is assumed that the visual classification collection of input video is combined into { C1,C2,C3,C4,C5,C6Totally 6
Class, using Si(Cx) indicate CxVisual classification coefficient, specifically, Si(Cx) indicate key frame KFiStatic classification result in be
The no visual classification C comprising in visual classification setx, by Si(Cx) it is used as CxVisual classification coefficient, wherein Cx∈{C1,C2,
C3,C4,C5,C6};As input video key KFiStatic classification result in include the visual classification C in visual classification setx, then
Si(Cx)=1;As input video key frame KFiStatic classification result in do not include visual classification set in visual classification Cx,
Then Si(Cx)=0.
For step S609, C in the visual classification set is calculatedxVisual classification weightAs shown in formula (8);
Wherein, wiFor KFiCorresponding video segmentation SiTime weighting, Si(Cx) indicate CxVisual classification coefficient.
For step S610, the default video classification parameter, the number for limiting input video classification results;
Wherein, the default video classification parameter needs are specifically determined according to actual conditions, for example, can be by input video
Visual classification set { C1,C2,C3,C4,C5,C6Corresponding each visual classification weightMiddle power
It is worth the first 3 larger final classification results as input video.
Present embodiments provide a kind of visual classification specific method, adjacent image frame in the system-computed input video
HS two-dimensional colors histogram and perceptual hash vector;The system is according to the HS two dimension face of adjacent image frame in the input video
Color Histogram and perceptual hash vector, calculate HS two-dimensional colors Histogram distance and the sense of adjacent image frame in the input video
Know Hash vector distance;When the HS two-dimensional color Histogram distances of adjacent image frame in the input video are more than preset first
Threshold value, and in the input video adjacent image frame perceptual hash vector distance be more than preset second threshold when, the system
System is segmented input video, obtains at least one video segmentation;The system obtains the average HS two dimensions of each video segmentation
Color histogram, and according to the average HS two-dimensional colors histogram of each video segmentation and average HS two-dimensional color histograms pair
The distance between the HS two-dimensional color histograms of the picture frame for the video segmentation answered relationship, obtains the corresponding pass of each video segmentation
Key frame;The system carries out image classification by Image Classifier to each key frame, is obtained according to default key frame classification parameter
The corresponding static classification set of each key frame;Wherein, described image grader is generated by deep neural network;The system
The number of image frames that the number of image frames and each video segmentation for including according to input video include calculates the time weighting of each video segmentation;
The system obtains the visual classification set of the input video according to the corresponding static classification set of each key frame;Institute
System is stated according to the relationship between each visual classification static classification set corresponding with each key frame in the visual classification set,
Obtain the visual classification coefficient of each visual classification in the visual classification set;The system according to each video segmentation when
Between in weight, the visual classification set each visual classification visual classification coefficient, calculate each video point in visual classification set
The visual classification weight of class;The system is according to the visual classification weight of each visual classification in the visual classification set and in advance
Setting video classification parameter obtains the final classification of input video as a result, carrying out visual classification using key frame solves visual classification
The problem of accuracy difference.
Embodiment three
Referring to Fig. 9, it illustrates a kind of structure of video classification system 90, the system comprises:First acquisition module
901, the second acquisition module 902, the first sort module 903 and the second sort module 904, wherein
First acquisition module 901, for obtaining at least one video segmentation from input video;
Second acquisition module 902, for according to the range performance in each video segmentation, obtaining each video segmentation and corresponding to
Key frame;
First sort module 903 obtains the corresponding static state of each key frame for carrying out image classification to each key frame
Classification set;
Second sort module 904, for according to the corresponding key frame of each video segmentation, the corresponding static state of each key frame
Classification set and default video classification parameter, obtain the classification results of the input video.
For first acquisition module 901, it is specifically used for
According between the HS two-dimensional colors Histogram distance and preset first threshold of adjacent image frame in input video
Corresponding pass in correspondence and input video between the perceptual hash vector distance of adjacent image frame and preset second threshold
System is segmented input video, obtains at least one video segmentation.
Further, first acquisition module 901, is specifically used for
Calculate the HS two-dimensional colors histogram of adjacent image frame and perceptual hash vector in input video;
And HS two-dimensional colors histogram and the perceptual hash vector according to adjacent image frame in the input video, meter
Calculate the HS two-dimensional colors Histogram distance and perceptual hash vector distance of adjacent image frame in the input video;
And when the HS two-dimensional color Histogram distances of adjacent image frame in the input video are more than preset first threshold
Value, and in the input video adjacent image frame perceptual hash vector distance be more than preset second threshold when, to input regard
Frequency is segmented, and at least one video segmentation is obtained.
For calculating the process of HS two-dimensional color histograms, it is with a picture frame in adjacent image frame in input video
Example, first acquisition module 901 include the first transformation submodule 9011 and statistic submodule 9012, and Figure 10 is the first acquisition mould
A kind of structure diagram of block 901, wherein
First transformation submodule 9011 is obtained for picture frame to be transformed to hsv color space from RGB color
Take H components, S components;
The statistic submodule 9012, for the corresponding channels H of H components channel S corresponding with S components to be divided into
A sections, statistics obtains the HS two-dimensional color histograms of picture frame.
Specifically, it for first transformation submodule 9011, is transformed into hsv color space from RGB color and obtains H
The formula of component and S components is respectively as shown in formula (1) and formula (2);
In formula (1) and formula (2), R indicates that the red component of input video, G indicate the green component of input video, B
Indicate that the blue component of input video, H indicate that the chrominance component of input video, S indicate the saturation degree component of input video, V tables
Show the lightness component of input video, V=max (R, G, B), 0≤V≤1;
Wherein, 0≤H≤360 in formula (1), if H<0, then H=H+360;In formula (2), 0≤S≤1.
For the statistic submodule 9012, it is preferable that first acquisition module is by the corresponding channels H of the H components
Channel S corresponding with S components is divided into 32 sections.
For the calculating process of perceptual hash vector, by taking a picture frame in adjacent image frame in input video as an example,
First acquisition module 901 further includes scaling submodule 9013, gray processing processing submodule 9014, the second transformation submodule
9015, submodule 9016, computational submodule 9017, comparison sub-module 9018 and vectorization submodule 9019 are chosen, Figure 11 the
Another structure diagram of one acquisition module 901, wherein
The scaling submodule 9013 obtains the first image array for picture frame to be zoomed to pre-set dimension;
The gray processing handles submodule 9014, for carrying out image gray processing processing to described first image matrix, obtains
Take the second image array;
Second transformation submodule 9015 obtains third image for carrying out two-dimensional dct to second image array
Matrix;
The selection submodule 9016 presets submatrix as the 4th image array for choosing in third image array;
The computational submodule 9017, the average pixel value for calculating the 4th image array;
The comparison sub-module 9018, for by the pixel value of each picture element in the 4th image array with it is described average
Pixel value is compared, and obtains newer 4th image array;
The vectorization submodule 9019 obtains final for newer 4th image array to be carried out vectorization
Perceptual hash vector.
For the scaling submodule 9013, it is preferable that picture frame is zoomed to the size of 32 × 32cm.
Specifically, for second transformation submodule 9015, shown in DCT formula such as formula (3);
Wherein, u, v=0,1,2 ..., N-1, it is the size value that picture frame is zoomed to pre-set dimension by the system, I to take N
(x, y) indicates image in the pixel value of (x, y) point, dct transform result of F (u, the v) expressions positioned at (u, v) point.
For the selection submodule 9016, it is preferable that choose the submatrix of 8 × 8cm of the upper left corner in third image array
As the 4th image array.
Specifically, for the comparison sub-module 9018, when the pixel value of the picture element in the 4th image array is more than institute
Average pixel value is stated, the pixel value of the picture element in the 4th image array is labeled as 0 by the system;When the 4th image moment
The pixel value of picture element in battle array is less than the average pixel value, and the system is by the picture element in the 4th image array
Pixel value is labeled as 1, to obtain newer 4th image array.
For first acquisition module 901, calculation formula such as formula (4) institute of the HS two-dimensional colors Histogram distance
Show;
Wherein, DHIndicate the two-dimensional color Histogram distance of adjacent image frame, Ht(x, y) indicates in adjacent image frame one
The HS two-dimensional colors histogram of picture frame is in the statistical value of (x, y) point, Ht+1(x, y) indicates another image in adjacent image frame
Statistical value of the HS two-dimensional colors histogram of frame in (x, y) point.
For first acquisition module 901, the calculation formula such as formula (5) and formula of the perceptual hash vector distance
(6) shown in;
In formula (5), DpIndicate perceptual hash vector distance, Pt(i) perception of a picture frame in adjacent image frame is indicated
I-th of element of Hash vector, Pt+1(i) vectorial i-th of the element of the perceptual hash of another picture frame in adjacent image frame is indicated;
In formula (6), P is worked as in D (x, x') expressionst+1(i)=Pt+1(i) when, D (x, x')=1;Work as Pt+1(i)≠Pt+1(i) when, D (x, x')
=0.
For the first acquisition module 901, it should be noted that when the adjacent image frame HS two-dimensional colors histogram away from
Perceptual hash vector distance and preset second from correspondence and adjacent image frame between preset first threshold
The HS two-dimensional color Histogram distances that correspondence between threshold value is unsatisfactory for the adjacent image frame are more than preset first threshold
Value, and the perceptual hash vector distance of adjacent image frame be more than preset second threshold when, first acquisition module 901 will not
Input video can be split.
For second acquisition module 902, it is specifically used for
The average HS two-dimensional colors histogram of each video segmentation is obtained, and according to each average HS two-dimensional colors histogram and respectively
Averagely the distance between HS two-dimensional color histograms of picture frame of the corresponding video segmentation of HS two-dimensional color histograms relationship, is obtained
Take the corresponding key frame of each video segmentation.
Further, second acquisition module 902, is specifically used for
By the average HS two-dimensional colors histogram of video segmentation video segmentation corresponding with average HS two-dimensional color histograms
In the HS two-dimensional color histograms of all picture frames be compared, the average HS two-dimensional colors of video segmentation described in selected distance
Key frame of the HS two-dimensional colors histogram of the nearest picture frame of histogram as the video segmentation.
For first sort module 903, it is specifically used for
Image classification is carried out to each key frame by Image Classifier, each key is obtained according to default key frame classification parameter
The corresponding static classification set of frame;Wherein, described image grader is generated by deep neural network.
Specifically, for the generation of Image Classifier, first sort module 903 is mainly used for
The other data representing image of each video class is collected, it, can also be by generation as the training data of disaggregated model
Table image data carries out the pretreatment operations such as image enhancement, rotation, random cropping, further training for promotion data bulk;
And training data is inputted into deep neural network, depth convolutional Neural net can be used in the deep neural network
The feature of the similar model extraction training datas of network such as AlexNet, GoogleNet or other is obtained by continuous repetitive exercise
Accurate image classification model is obtained as Image Classifier.
To first sort module 903, the default key frame classification parameter is corresponding quiet for limiting each key frame
The number of crucial frame category in state classification set;
Wherein, the default key frame classification parameter needs are specifically determined according to actual conditions, for example, by successive ignition
Trained deep neural network, can be using the larger preceding k classes of probability in the classification results of deep neural network output as input
The static classification result of key frame of video.
As shown in figure 12, second sort module 904 includes the first acquisition submodule 9041, the second acquisition submodule
9042, third acquisition submodule 9043, the 4th acquisition submodule 9044 and the 5th acquisition submodule 9045, wherein
First acquisition submodule 9041, number of image frames and each video segmentation for including according to input video include
Number of image frames calculate the time weighting of each video segmentation;
Second acquisition submodule 9042, for according to the corresponding static classification set of each key frame, obtaining institute
State the visual classification set of input video;
The third acquisition submodule 9043, for according to each visual classification in the visual classification set and each key frame
Relationship between corresponding static classification set obtains the visual classification coefficient of each visual classification in the visual classification set;
4th acquisition submodule 9044, for time weighting, the visual classification according to each video segmentation
The visual classification coefficient of each visual classification in set calculates the visual classification weight of each visual classification in visual classification set;
5th acquisition submodule 9045, for the visual classification according to each visual classification in the visual classification set
Weight and default video classification parameter, obtain the final classification result of input video.
For first acquisition submodule 9041, it is assumed that input video is made of N number of picture frame, wherein key frame KF1,
KF2,...,KFnMiddle KFiCorresponding video segmentation SiBy niA picture frame composition, KFiCorresponding video segmentation SiTime weighting wi
Shown in calculation formula such as formula (7).
For second acquisition submodule 9042, specifically, by the corresponding static classification union of sets collection of each key frame
Visual classification set as the input video.
For the third acquisition submodule 9043, Si(Cx) indicate CxVisual classification coefficient, specifically, Si(Cx) indicate
Key frame KFiStatic classification result in whether include the visual classification C in visual classification setx;As input video key KFi
Static classification result in include the visual classification C in visual classification setx, then Si(Cx)=1;As input video key frame KFi
Static classification result in do not include visual classification set in visual classification Cx, then Si(Cx)=0.
For the 4th acquisition submodule 9044, C in the visual classification set is calculatedxVisual classification weight WCxSuch as
Shown in formula (8);
Wherein, wiFor KFiCorresponding video segmentation SiTime weighting, Si(Cx) indicate CxVisual classification coefficient.
For the 5th acquisition submodule 9045, the default video classification parameter, for limiting input video classification
As a result number;
Wherein, the default video classification parameter needs are specifically determined according to actual conditions, for example, can be by input video
The corresponding visual classification weight of visual classification set in the larger first k' final classification result as input video of weights.
Specifically, for the present embodiment, first acquisition module 901, the second acquisition module 902, the first sort module
903 and second the function of sort module 904 program or pre- storage in memory can be called by the processor of the system 90
Data are realized.In practical applications, above-mentioned processor can be application-specific IC (ASIC, Application
Specific Integrated Circuit), digital signal processor (DSP, Digital Signal Processor), number
Word signal processing apparatus (DSPD, Digital Signal Processing Device), programmable logic device (PLD,
Programmable Logic Device), field programmable gate array (FPGA, Field Programmable Gate
Array), in central processing unit (CPU, Central Processing Unit), controller, microcontroller, microprocessor extremely
Few one kind.It is to be appreciated that for different systems, the electronic device for realizing above-mentioned processor function can also be it
It, the embodiment of the present invention is not especially limited.
Present embodiments provide a kind of video classification system, first acquisition module 901, for being obtained from input video
Take at least one video segmentation;Second acquisition module 902, for according to the range performance in each video segmentation, obtaining each
The corresponding key frame of video segmentation;First sort module 903 obtains each pass for carrying out image classification to each key frame
The corresponding static classification set of key frame;Second sort module 904 is used for according to the corresponding key frame of each video segmentation, respectively
The corresponding static classification set of key frame and default video classification parameter, obtain the classification results of the input video, using pass
Key frame carries out visual classification and solves the problems, such as that visual classification accuracy is poor.
It should be understood by those skilled in the art that, the embodiment of the present invention can be provided as method, system or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the present invention
Formula.Moreover, the present invention can be used can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The present invention be with reference to according to the method for the embodiment of the present invention, the flow of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that can be realized by computer program instructions every first-class in flowchart and/or the block diagram
The combination of flow and/or box in journey and/or box and flowchart and/or the block diagram.These computer programs can be provided
Instruct the processor of all-purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine so that the instruction executed by computer or the processor of other programmable data processing devices is generated for real
The device for the function of being specified in present one flow of flow chart or one box of multiple flows and/or block diagram or multiple boxes.
These computer program instructions, which may also be stored in, can guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works so that instruction generation stored in the computer readable memory includes referring to
Enable the manufacture of device, the command device realize in one flow of flow chart or multiple flows and/or one box of block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device so that count
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, in computer or
The instruction executed on other programmable devices is provided for realizing in one flow of flow chart or multiple flows and/or block diagram one
The step of function of being specified in a box or multiple boxes.
The foregoing is only a preferred embodiment of the present invention, is not intended to limit the scope of the present invention.
Claims (26)
1. a kind of video classification methods, which is characterized in that the method is used for a kind of video classification system, the method includes:
The system obtains at least one video segmentation from input video;
The system obtains the corresponding key frame of each video segmentation according to the range performance in each video segmentation;
The system carries out image classification to each key frame, obtains the corresponding static classification set of each key frame;
The system is according to the corresponding key frame of each video segmentation, the corresponding static classification set of each key frame and default video class
Other parameter, obtains the classification results of the input video.
2. according to the method described in claim 1, it is characterized in that, the system obtains at least one video from input video
Segmentation, specifically includes:
The system according to the HS two-dimensional colors Histogram distance of adjacent image frame in input video and preset first threshold it
Between correspondence and input video in adjacent image frame perceptual hash vector distance and preset second threshold between pair
It should be related to and input video is segmented, obtain at least one video segmentation.
3. according to the method described in claim 2, it is characterized in that, the system is according to the HS of adjacent image frame in input video
The perception of adjacent image frame in correspondence and input video between two-dimensional color Histogram distance and preset first threshold
Correspondence between Hash vector distance and preset second threshold is segmented input video, obtains at least one video
Segmentation, specifically includes:
In the system-computed input video tone saturation degree HS two-dimensional colors histogram of adjacent image frame and perceptual hash to
Amount;
The system is calculated according to HS two-dimensional colors histogram and the perceptual hash vector of adjacent image frame in the input video
The HS two-dimensional colors Histogram distance and perceptual hash vector distance of adjacent image frame in the input video;
It is and described when the HS two-dimensional color Histogram distances of adjacent image frame in the input video are more than preset first threshold
When the perceptual hash vector distance of adjacent image frame is more than preset second threshold in input video, the system is to input video
It is segmented, obtains at least one video segmentation.
4. according to the method described in claim 3, it is characterized in that, the process tool of the system-computed HS two-dimensional color histograms
Body includes:
Picture frame is transformed into tone saturation degree lightness hsv color space by the system from RGB RGB color, obtains color
Adjust H components, saturation degree S components;
The corresponding channels H of H components channel S corresponding with S components is divided into a sections by the system, and statistics obtains picture frame
HS two-dimensional color histograms.
5. according to the method described in claim 3, it is characterized in that, the process of the system-computed perceptual hash vector is specifically wrapped
It includes:
Picture frame is zoomed to pre-set dimension by the system, obtains the first image array;
The system carries out image gray processing processing to described first image matrix, obtains the second image array;
The system carries out two-dimension discrete cosine transform DCT to second image array, obtains third image array;
The system, which is chosen, presets submatrix as the 4th image array in third image array;
The average pixel value of 4th image array described in the system-computed;
The pixel value of each picture element in 4th image array is compared by the system with the average pixel value, is obtained
Newer 4th image array;
Newer 4th image array is carried out vectorization by the system, obtains final perceptual hash vector.
6. according to the method described in claim 2, it is characterized in that, when the adjacent image frame HS two-dimensional colors histogram away from
Perceptual hash vector distance and preset second from correspondence and adjacent image frame between preset first threshold
The HS two-dimensional color Histogram distances that correspondence between threshold value is unsatisfactory for the adjacent image frame are more than preset first threshold
Value, and the perceptual hash vector distance of adjacent image frame be more than preset second threshold when, the system will not to input regard
Frequency is split.
7. according to the method described in claim 1, it is characterized in that, the system according to the range performance in each video segmentation,
The corresponding key frame of each video segmentation is obtained, is specifically included:
The system obtains the average HS two-dimensional colors histogram of each video segmentation, and according to each average HS two-dimensional color histograms
The distance between the HS two-dimensional color histograms of picture frame of video segmentation corresponding with each averagely HS two-dimensional color histograms close
System, obtains the corresponding key frame of each video segmentation.
8. the method according to the description of claim 7 is characterized in that the system is straight by the average HS two-dimensional colors of video segmentation
The HS two-dimensional color histograms of all picture frames carry out in side's figure video segmentation corresponding with average HS two-dimensional color histograms
Compare, the HS two-dimensional color histograms of the average nearest picture frame of HS two-dimensional colors histogram of video segmentation described in selected distance
Key frame as the video segmentation.
9. according to the method described in claim 1, it is characterized in that, the system carries out image classification, acquisition to each key frame
The corresponding static classification set of each key frame, specifically includes:
The system carries out image classification by Image Classifier to each key frame, is obtained according to default key frame classification parameter each
The corresponding static classification set of key frame;Wherein, described image grader is generated by deep neural network.
10. according to the method described in claim 9, it is characterized in that, the default key frame classification parameter, for limiting each pass
The number of crucial frame category in the corresponding static classification set of key frame.
11. according to the method described in claim 1, it is characterized in that, the system according to the corresponding key frame of each video segmentation,
Each corresponding static classification set of key frame and default video classification parameter, obtain the classification results of the input video, specifically
Including:
The number of image frames that the number of image frames and each video segmentation that the system includes according to input video include calculates each video point
The time weighting of section;
The system obtains the visual classification collection of the input video according to the corresponding static classification set of each key frame
It closes;
The system is according between each visual classification static classification set corresponding with each key frame in the visual classification set
Relationship, obtain the visual classification coefficient of each visual classification in the visual classification set;
The system is according to the video point of each visual classification in the time weighting of each video segmentation, the visual classification set
Class coefficient calculates the visual classification weight of each visual classification in visual classification set;
The system is joined according to the visual classification weight of each visual classification in the visual classification set and default video classification
Number, obtains the final classification result of input video.
12. according to the method for claim 11, which is characterized in that the system is by the corresponding static classification collection of each key frame
Visual classification set of the union of conjunction as the input video.
13. according to the method for claim 11, which is characterized in that the default video classification parameter, for limiting input
The number of visual classification result.
14. a kind of video classification system, which is characterized in that the system comprises:First acquisition module, the second acquisition module,
One sort module and the second sort module, wherein
First acquisition module, for obtaining at least one video segmentation from input video;
Second acquisition module, for according to the range performance in each video segmentation, obtaining the corresponding key of each video segmentation
Frame;
First sort module obtains the corresponding static classification collection of each key frame for carrying out image classification to each key frame
It closes;
Second sort module, for according to the corresponding key frame of each video segmentation, the corresponding static classification collection of each key frame
Conjunction and default video classification parameter, obtain the classification results of the input video.
15. system according to claim 14, which is characterized in that first acquisition module is specifically used for
According to corresponding between the HS two-dimensional colors Histogram distance of adjacent image frame in input video and preset first threshold
Correspondence pair in relationship and input video between the perceptual hash vector distance and preset second threshold of adjacent image frame
Input video is segmented, and at least one video segmentation is obtained.
16. system according to claim 15, which is characterized in that first acquisition module is specifically used for
Calculate the HS two-dimensional colors histogram of adjacent image frame and perceptual hash vector in input video;
According to the HS two-dimensional colors histogram of adjacent image frame in the input video and perceptual hash vector, the input is calculated
The HS two-dimensional colors Histogram distance and perceptual hash vector distance of adjacent image frame in video;
It is and described when the HS two-dimensional color Histogram distances of adjacent image frame in the input video are more than preset first threshold
When the perceptual hash vector distance of adjacent image frame is more than preset second threshold in input video, input video is divided
Section, obtains at least one video segmentation.
17. system according to claim 16, which is characterized in that for calculating the process of HS two-dimensional color histograms, institute
It includes the first transformation submodule and statistic submodule to state the first acquisition module, wherein
First transformation submodule obtains H components, S for picture frame to be transformed into hsv color space from RGB color
Component;
The statistic submodule, for the corresponding channels H of H components channel S corresponding with S components to be divided into a sections, statistics
Obtain the HS two-dimensional color histograms of picture frame.
18. system according to claim 16, which is characterized in that for calculating the process of perceptual hash vector, described the
First acquisition module described in one acquisition module further includes scaling submodule, gray processing processing submodule, the second transformation submodule, choosing
Take submodule, computational submodule, comparison sub-module and vectorization submodule, wherein
The scaling submodule obtains the first image array for picture frame to be zoomed to pre-set dimension;
The gray processing handles submodule, for carrying out image gray processing processing to described first image matrix, obtains the second figure
As matrix;
Second transformation submodule obtains third image array for carrying out two-dimensional dct to second image array;
The selection submodule presets submatrix as the 4th image array for choosing in third image array;
The computational submodule, the average pixel value for calculating the 4th image array;
The comparison sub-module, for by the pixel value of each picture element in the 4th image array and the average pixel value into
Row compares, and obtains newer 4th image array;
The vectorization submodule obtains final perception and breathes out for newer 4th image array to be carried out vectorization
Uncommon vector.
19. system according to claim 15, which is characterized in that when the HS two-dimensional color histograms of the adjacent image frame
The perceptual hash vector distance of correspondence and adjacent image frame between distance and preset first threshold and preset the
The HS two-dimensional color Histogram distances that correspondence between two threshold values is unsatisfactory for the adjacent image frame are more than preset first
Threshold value, and the perceptual hash vector distance of adjacent image frame be more than preset second threshold when, first acquisition module will not
Input video can be split.
20. system according to claim 14, which is characterized in that second acquisition module is specifically used for
Obtain the average HS two-dimensional colors histogram of each video segmentation, and according to each average HS two-dimensional colors histogram with it is each average
The distance between the HS two-dimensional color histograms of the picture frame of the corresponding video segmentation of HS two-dimensional color histograms relationship obtains each
The corresponding key frame of video segmentation.
21. system according to claim 20, which is characterized in that second acquisition module is specifically used for
By institute in the average HS two-dimensional colors histogram of video segmentation video segmentation corresponding with average HS two-dimensional color histograms
The HS two-dimensional color histograms of some picture frames are compared, the average HS two-dimensional colors histogram of video segmentation described in selected distance
Scheme key frame of the HS two-dimensional colors histogram of nearest picture frame as the video segmentation.
22. system according to claim 14, which is characterized in that first sort module is specifically used for
Image classification is carried out to each key frame by Image Classifier, each key frame pair is obtained according to default key frame classification parameter
The static classification set answered;Wherein, described image grader is generated by deep neural network.
23. system according to claim 22, which is characterized in that the default key frame classification parameter, it is each for limiting
The number of crucial frame category in the corresponding static classification set of key frame.
24. system according to claim 14, which is characterized in that second sort module includes the first acquisition submodule
Block, the second acquisition submodule, third acquisition submodule, the 4th acquisition submodule and the 5th acquisition submodule, wherein
First acquisition submodule, the picture frame that number of image frames and each video segmentation for including according to input video include
Number calculates the time weighting of each video segmentation;
Second acquisition submodule, for according to the corresponding static classification set of each key frame, obtaining the input and regarding
The visual classification set of frequency;
The third acquisition submodule, for corresponding with each key frame quiet according to each visual classification in the visual classification set
Relationship between state classification set, obtains the visual classification coefficient of each visual classification in the visual classification set;
4th acquisition submodule, for according to each in the time weighting of each video segmentation, the visual classification set
The visual classification coefficient of visual classification calculates the visual classification weight of each visual classification in visual classification set;
5th acquisition submodule, for according to the visual classification weight of each visual classification in the visual classification set and
Default video classification parameter, obtains the final classification result of input video.
25. according to the method for claim 24, which is characterized in that second acquisition submodule is specifically used for
Using the corresponding static classification union of sets collection of each key frame as the visual classification set of the input video.
26. according to the method for claim 24, which is characterized in that in the 5th acquisition submodule, the pre- setting video
Classification parameter, the number for limiting input video classification results.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611137674.1A CN108615043B (en) | 2016-12-12 | 2016-12-12 | Video classification method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611137674.1A CN108615043B (en) | 2016-12-12 | 2016-12-12 | Video classification method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108615043A true CN108615043A (en) | 2018-10-02 |
CN108615043B CN108615043B (en) | 2021-05-25 |
Family
ID=63643854
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611137674.1A Active CN108615043B (en) | 2016-12-12 | 2016-12-12 | Video classification method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108615043B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109840509A (en) * | 2019-02-15 | 2019-06-04 | 北京工业大学 | The multi-level collaboration recognition methods of bad main broadcaster and device in network direct broadcasting video |
CN110414335A (en) * | 2019-06-20 | 2019-11-05 | 北京奇艺世纪科技有限公司 | Video frequency identifying method, device and computer readable storage medium |
WO2020107625A1 (en) * | 2018-11-27 | 2020-06-04 | 北京微播视界科技有限公司 | Video classification method and apparatus, electronic device, and computer readable storage medium |
CN112380954A (en) * | 2020-11-10 | 2021-02-19 | 四川长虹电器股份有限公司 | Video classification intercepting system and method based on image recognition |
CN112861609A (en) * | 2020-12-30 | 2021-05-28 | 中国电子科技集团公司信息科学研究院 | Method for improving multi-thread content key frame identification efficiency |
CN115205768A (en) * | 2022-09-16 | 2022-10-18 | 山东百盟信息技术有限公司 | Video classification method based on resolution self-adaptive network |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6381363B1 (en) * | 1999-03-15 | 2002-04-30 | Grass Valley (U.S.), Inc. | Histogram-based segmentation of images and video via color moments |
US20070183497A1 (en) * | 2006-02-03 | 2007-08-09 | Jiebo Luo | Extracting key frame candidates from video clip |
CN101604325A (en) * | 2009-07-17 | 2009-12-16 | 北京邮电大学 | Method for classifying sports video based on key frame of main scene lens |
CN102663015A (en) * | 2012-03-21 | 2012-09-12 | 上海大学 | Video semantic labeling method based on characteristics bag models and supervised learning |
US20120237126A1 (en) * | 2011-03-16 | 2012-09-20 | Electronics & Telecommunications Research Institute | Apparatus and method for determining characteristic of motion picture |
CN103593464A (en) * | 2013-11-25 | 2014-02-19 | 华中科技大学 | Video fingerprint detecting and video sequence matching method and system based on visual features |
US20140210944A1 (en) * | 2013-01-30 | 2014-07-31 | Samsung Electronics Co., Ltd. | Method and apparatus for converting 2d video to 3d video |
CN104123396A (en) * | 2014-08-15 | 2014-10-29 | 三星电子(中国)研发中心 | Soccer video abstract generation method and device based on cloud television |
CN105069042A (en) * | 2015-07-23 | 2015-11-18 | 北京航空航天大学 | Content-based data retrieval methods for unmanned aerial vehicle spying images |
CN106156284A (en) * | 2016-06-24 | 2016-11-23 | 合肥工业大学 | Video retrieval method is closely repeated based on random the extensive of various visual angles Hash |
-
2016
- 2016-12-12 CN CN201611137674.1A patent/CN108615043B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6381363B1 (en) * | 1999-03-15 | 2002-04-30 | Grass Valley (U.S.), Inc. | Histogram-based segmentation of images and video via color moments |
US20070183497A1 (en) * | 2006-02-03 | 2007-08-09 | Jiebo Luo | Extracting key frame candidates from video clip |
CN101604325A (en) * | 2009-07-17 | 2009-12-16 | 北京邮电大学 | Method for classifying sports video based on key frame of main scene lens |
US20120237126A1 (en) * | 2011-03-16 | 2012-09-20 | Electronics & Telecommunications Research Institute | Apparatus and method for determining characteristic of motion picture |
CN102663015A (en) * | 2012-03-21 | 2012-09-12 | 上海大学 | Video semantic labeling method based on characteristics bag models and supervised learning |
US20140210944A1 (en) * | 2013-01-30 | 2014-07-31 | Samsung Electronics Co., Ltd. | Method and apparatus for converting 2d video to 3d video |
CN103593464A (en) * | 2013-11-25 | 2014-02-19 | 华中科技大学 | Video fingerprint detecting and video sequence matching method and system based on visual features |
CN104123396A (en) * | 2014-08-15 | 2014-10-29 | 三星电子(中国)研发中心 | Soccer video abstract generation method and device based on cloud television |
CN105069042A (en) * | 2015-07-23 | 2015-11-18 | 北京航空航天大学 | Content-based data retrieval methods for unmanned aerial vehicle spying images |
CN106156284A (en) * | 2016-06-24 | 2016-11-23 | 合肥工业大学 | Video retrieval method is closely repeated based on random the extensive of various visual angles Hash |
Non-Patent Citations (3)
Title |
---|
SHIYAN HU: ""Efficient video retrieval by locality sensitive hashing"", 《PROCEEDINGS. (ICASSP "05). IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, 2005》 * |
YIFENG LIU等: ""Segmentation by weighted aggregation and perceptual hash for pedestrian detection"", 《JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION》 * |
西华大学: ""结合裁剪技术的颜色特征匹配搜索"", 《宜宾学院学报》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020107625A1 (en) * | 2018-11-27 | 2020-06-04 | 北京微播视界科技有限公司 | Video classification method and apparatus, electronic device, and computer readable storage medium |
CN109840509A (en) * | 2019-02-15 | 2019-06-04 | 北京工业大学 | The multi-level collaboration recognition methods of bad main broadcaster and device in network direct broadcasting video |
CN110414335A (en) * | 2019-06-20 | 2019-11-05 | 北京奇艺世纪科技有限公司 | Video frequency identifying method, device and computer readable storage medium |
CN112380954A (en) * | 2020-11-10 | 2021-02-19 | 四川长虹电器股份有限公司 | Video classification intercepting system and method based on image recognition |
CN112861609A (en) * | 2020-12-30 | 2021-05-28 | 中国电子科技集团公司信息科学研究院 | Method for improving multi-thread content key frame identification efficiency |
CN112861609B (en) * | 2020-12-30 | 2024-04-09 | 中国电子科技集团公司信息科学研究院 | Multithreading content key frame identification efficiency improvement method |
CN115205768A (en) * | 2022-09-16 | 2022-10-18 | 山东百盟信息技术有限公司 | Video classification method based on resolution self-adaptive network |
Also Published As
Publication number | Publication date |
---|---|
CN108615043B (en) | 2021-05-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108615043A (en) | A kind of video classification methods and system | |
CN107767413B (en) | Image depth estimation method based on convolutional neural network | |
Li et al. | Investigation on cost assignment in spatial image steganography | |
CN109493350A (en) | Portrait dividing method and device | |
EP1280107A2 (en) | Quality based image compression | |
CN106851437A (en) | A kind of method for extracting video frequency abstract | |
Rathore et al. | Colour based image segmentation using L* a* b* colour space based on genetic algorithm | |
Diana Andrushia et al. | Saliency-based image compression using Walsh–Hadamard transform (WHT) | |
CN110136166B (en) | Automatic tracking method for multi-channel pictures | |
CN107481236A (en) | A kind of quality evaluating method of screen picture | |
CN110136083B (en) | Base map updating method and device combined with interaction | |
CN110415207A (en) | A method of the image quality measure based on image fault type | |
CN113301408A (en) | Video data processing method and device, electronic equipment and readable storage medium | |
CN111179370B (en) | Picture generation method and device, electronic equipment and storage medium | |
JP6387026B2 (en) | Book searching apparatus, method and program | |
CN111641822A (en) | Method for evaluating quality of repositioning stereo image | |
Miyata et al. | Novel inverse colorization for image compression | |
CN105678718A (en) | Method and device for image denoising | |
US6510242B1 (en) | Color image enhancement during YCbCr to RGB color conversion | |
KR100340035B1 (en) | Image similarity weight adjustment apparatus and method, and content-based image search system and method using same | |
CN108230411B (en) | Method and device for detecting tampered image | |
CN104574320B (en) | A kind of image super-resolution restored method based on sparse coding coefficients match | |
CN110647898B (en) | Image processing method, image processing device, electronic equipment and computer storage medium | |
CN105989029A (en) | Image searching method and image searching system | |
CN113395407A (en) | Image processing apparatus, image processing method, and computer readable medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information |
Address after: 310012 building A01, 1600 yuhangtang Road, Wuchang Street, Yuhang District, Hangzhou City, Zhejiang Province Applicant after: CHINA MOBILE (HANGZHOU) INFORMATION TECHNOLOGY Co.,Ltd. Applicant after: China Mobile Communications Corp. Address before: 310012, No. 14, building three, Chang Torch Hotel, No. 259, Wensanlu Road, Xihu District, Zhejiang, Hangzhou Applicant before: CHINA MOBILE (HANGZHOU) INFORMATION TECHNOLOGY Co.,Ltd. Applicant before: China Mobile Communications Corp. |
|
CB02 | Change of applicant information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |