CN106326927B - A kind of shoes print new category detection method - Google Patents
A kind of shoes print new category detection method Download PDFInfo
- Publication number
- CN106326927B CN106326927B CN201610716111.1A CN201610716111A CN106326927B CN 106326927 B CN106326927 B CN 106326927B CN 201610716111 A CN201610716111 A CN 201610716111A CN 106326927 B CN106326927 B CN 106326927B
- Authority
- CN
- China
- Prior art keywords
- image
- detected
- similarity
- training image
- training
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of shoes print new category detection method, comprising: extracts the feature of training image and described image to be detected, the training image is the shoes watermark image of known class, for determining whether image to be detected is new category shoes watermark image；Extract the feature of the training image and described image to be detected；According to the similarity matrix between similarity matrix, the training image and described image to be detected between training image described in the feature calculation；Discriminant function is determined according to the similarity matrix between the training image；Determine the training image and image to be detected in the corresponding training image mapping of kernel and image to be detected mapping according to the discriminant function；Calculate the Euclidean distance of image to be detected mapping and training image mapping；Determine whether image to be detected is new category according to the Euclidean distance；Shoes watermark image is detected according to the new category.The present invention realizes the effective management printed to shoes, realizes more accurately detection shoes print new category.
Description
Technical field
The present embodiments relate to art of image analysis more particularly to a kind of shoes to print new category detection method.
Background technique
There is the new category detection algorithm based on shoes print data set, human face data collection and general data collection at present, is printed based on shoes
New category detection algorithm have using the cascade opener mark image classification method of multilayer；New category detection algorithm based on face
Have: based on similarity distribution opener face recognition algorithms, in conjunction with the Adaboost opener face recognition algorithms of geometric transformation, base
In the opener face recognition algorithms that transduction principle and KNN are combined；Having in conjunction with kernel and core Foley- based on general data collection
The new category detection algorithm of Sammon transformation.The main thought of each method is as follows: (1) the new category detection algorithm based on shoes print is
Using the cascade opener mark image classification method of multilayer, this method is first according to pretreated image to be classified and trace figure
As the candidate categories of the similitude screening image to be classified of mark image each in library, image to be classified and candidate class are then calculated
Rank the first the mark image of position in not and its similitude of corresponding representative image recycle the cascade mode of multilayer judge to point
Whether class image belongs to a certain classification or new category image in mark image library.(2) the new category detection algorithm based on face
There are several types of: the opener face recognition algorithms based on similarity distribution, this method pass through the test specimens of a large amount of tape identification first
The similarity vector of this acquisition sample, and classify to test sample according to following three kinds of situations: 1) test sample belongs to known class
Not, and classify correct；2) test sample belongs to known class, but classification error；3) test sample is not belonging to known class.By
A kind of vector of situation is classified as 0 class, and the vector of the two or three kind of situation is classified as 1 class.Then it introduces linear discriminant and finds out optimal differentiation
Hyperplane judges whether test sample belongs to using the hyperplane learnt after a new test sample enters identifying system
Known class；In conjunction with the Adaboost opener face recognition algorithms of geometric transformation, this method is using classifier to the tendency of positive sample
Property, so that the new samples that former positive sample transformation generates still can be by classifier, the approximate sample of negative sample is but not easily passed through, energy
The positive negative sample of overlapping region is effectively distinguished, and reduces the expense of time by two layers of identification structure, makes classifier correct
In the case that discrimination is constant, false acceptance rate can be greatly reduced；The opener recognition of face combined based on transduction principle and KNN
Algorithm, this method combined first with transduction principle and KNN calculate P value, that is, test sample one of test sample effectively with
Machine test, by the corresponding label output of maximum P value if maximum P value is significantly more than other P values, if maximum P value with
Neighbouring score, which is not much different but differs greatly with other P values, then still exports the corresponding label of maximum P value, only sets
Reliability is very low, if all P value random distributions and not obviously should be to this sample rejection if the P value more much larger than other values
Not.It (3) based on general data collection is a kind of new category detection method converted in conjunction with kernel and core Foley-Sammon, it should
Method calculates image to be detected mapping and training image by the way that training image and image to be detected are respectively mapped to kernel
Then the Euclidean distance of mapping is realized to be detected by the smallest distance as final new category score according to the threshold value of setting
Collection is outer in the collection of image determines.
Above-mentioned shoes print new category detection algorithm there are the problem of: (1) asked existing for the new category detection algorithm based on shoes print
Topic is that different addition strategies is taken different score sections based on the cascade opener mark image classification method of multilayer, and threshold
The setting of value is very crucial, needs very big skill.(2) the new category detection algorithm based on face the problem is that: be based on phase
Differentiated like the opener recognition methods of degree using the entire similarity vector of test sample, rather than considers K neighbour's sample
Similarity, there is no by similarity value it is especially low sample rejecting may will affect differentiation effect in this way；In conjunction with geometric transformation
Adaboost opener identification need transformed n sample passing sequentially through classifier, see that can transformed sample intensive
Pass through, it is done so that increase of the expense on recognition time at n times, be not suitable for the knowledge containing a large amount of face databases
Not；Need first to calculate test sample and all trained classifications based on transduction principle and KNN the opener face recognition algorithms combined
Unqualified degree calculates P value again according to unqualified degree to form the forecast set of test sample, and the distribution determination further according to P value is refused
The threshold value of identification, the computation complexity of this method is high, memory consumption is big, therefore is not suitable for the number containing a large amount of training sample classifications
According to collection.(3) the new category detection algorithm based on general data collection the problem is that: if only using arest neighbors classification one dimension
The range information of degree carries out new category detection, will lose the range information with other classifications, influences the efficiency differentiated.To sum up institute
It states, the Detection accuracy that prior technical shoes print new category is low.
Summary of the invention
The embodiment of the present invention provides a kind of shoes print new category detection method, to overcome above-mentioned technical problem.
A kind of shoes of the present invention print new category detection method, comprising:
The feature of the training image and described image to be detected is extracted, the training image is the shoes impression of known class
Picture, for determining whether image to be detected is new category shoes watermark image；
According between training image described in the feature calculation similarity matrix, the training image with it is described to be detected
Similarity matrix between image；
Discriminant function is determined according to the similarity matrix between the training image；
Determine that the training image and image to be detected are reflected in the corresponding training image of kernel according to the discriminant function
It penetrates and is mapped with image to be detected；
Calculate the Euclidean distance of image to be detected mapping and training image mapping；
Determine whether image to be detected is new category according to the Euclidean distance.
Further, the feature for extracting the training image and described image to be detected, comprising:
It will be used as sole above the training image and image to be detected according to preset value, remainder is as heel；
By the sole, heel image mirrors processing；
Sole, heel image after original image and mirror image is subjected to a quarter that wavelet transformation is the original image；
Image after the wavelet transformation is subjected to polar coordinate transform, and extracts Fourier transformation and obtains feature.
Further, the similarity matrix between the training image according to the feature calculation, the training figure
Picture and the similarity matrix between described image to be detected, comprising:
According to the similarity between the sole before and after training image mirror image processing described in two width of feature calculation；And compare
Two similarities, by the biggish similarity as the two width training image sole of similarity；
According to the similarity between the heel before and after training image mirror image processing described in two width of feature calculation；And compare
Two similarities, by the biggish similarity as the two width training image heel of similarity；
According to the similarity between the sole before and after the feature calculation training image and image to be detected mirror image processing；And
Compare two similarities, by similarity it is biggish as between the training image and described image to be detected sole it is similar
Degree；
According to the similarity between the heel before and after the feature calculation training image and image to be detected mirror image processing；And
Compare two similarities, by similarity it is biggish as between the training image and described image to be detected heel it is similar
Degree.
It is further, described that discriminant function is determined according to the similarity matrix, comprising:
The similarity matrix of the training image is subjected to centralization；
The characteristic value and feature vector of similarity matrix after seeking the centralization, and give up the characteristic value less than zero pair
The feature vector answered；
According to described eigenvector computational discrimination function.
Further, described to determine that the training image and image to be detected are corresponding in kernel according to the discriminant function
Training image mapping and image to be detected mapping, comprising:
According to the discriminant function training image of the same category is mapped to kernel one by one and obtains a kind of training image
Mapping；
The similarity for calculating image to be detected and all training images obtains similarity vector, and by the similarity vector
Each component sorts from large to small, and image to be detected mapping is obtained after preceding K similarity component is multiplied with the discriminant function.
It is further, described to determine whether image to be detected is new category according to the Euclidean distance, comprising:
By Euclidean distance according to sequence from small to large, and sequence first place and last N of ratio are calculated separately, by institute
The Euclidean distance value summation for stating N number of ratio and the sequence first place determines new category value；
By the new category value and threshold value comparison, determine that described image to be detected is if the new category value is greater than threshold value
New category determines that described image to be detected belongs to existing classification if the new category value is less than the threshold value.
The present invention realizes the effective management printed to shoes, realizes more accurately detection shoes print new category, improves work effect
Rate.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without any creative labor, can be with
It obtains other drawings based on these drawings.
Fig. 1 is that shoes of the present invention print new category detection method flow chart；
Fig. 2 is that shoes of the present invention print new category detection method overall flow figure；
Fig. 3 is that image to be detected of the present invention belongs to schematic diagram in collection；
Fig. 4 is that image to be detected of the present invention belongs to the outer schematic diagram of collection.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Fig. 1 is that shoes of the present invention print detection method flow chart, as shown in Figure 1, the method for the present embodiment may include:
Step 101, the feature for extracting the training image and described image to be detected, the training image are known class
Shoes watermark image, for determining whether image to be detected is new category shoes watermark image；
Similarity matrix, the training image and institute between step 102, the training image according to the feature calculation
State the similarity matrix between image to be detected；
Step 103 determines discriminant function according to the similarity matrix between the training image；
Step 104 determines the training image and image to be detected in the corresponding instruction of kernel according to the discriminant function
Practice image mapping and image to be detected mapping；
Step 105, the Euclidean distance for calculating image to be detected mapping and training image mapping；
Step 106 determines whether image to be detected is new category according to the Euclidean distance.
Further, the feature for extracting the training image and described image to be detected, comprising:
It will be used as sole above the training image and image to be detected according to preset value, remainder is as heel；
By the sole, heel image mirrors processing；
Sole, heel image after original image and mirror image is subjected to a quarter that wavelet transformation is the original image；
Image after the wavelet transformation is subjected to polar coordinate transform, and extracts Fourier transformation and obtains feature.
Further, the similarity matrix between the training image according to the feature calculation, the training figure
Picture and the similarity matrix between described image to be detected, comprising:
According to the similarity between the sole before and after training image mirror image processing described in two width of feature calculation；And compare
Two similarities, by the biggish similarity as the two width training image sole of similarity；
According to the similarity between the heel before and after training image mirror image processing described in two width of feature calculation；And compare
Two similarities, by the biggish similarity as the two width training image heel of similarity；
According to the similarity between the sole before and after the feature calculation training image and image to be detected mirror image processing；And
Compare two similarities, by similarity it is biggish as between the training image and described image to be detected sole it is similar
Degree；
According to the similarity between the heel before and after the feature calculation training image and image to be detected mirror image processing；And
Compare two similarities, by similarity it is biggish as between the training image and described image to be detected heel it is similar
Degree.
Specifically, using above image 60 percent shoes watermark image as sole, residue percent in the present embodiment
For 40 shoes watermark image as heel, then by sole, heel mirror image, then carrying out wavelet transformation to each section becomes original image
Then a quarter carries out polar coordinate transform and takes Fourier transformation to obtain feature again, using this tetrameric feature as a width figure
The global feature of picture.The similarity between similarity, sole and the sole mirror image between corresponding sole is calculated, in order to eliminate left and right
The influence of foot, the similarity as two images sole for taking similarity big；Similarity between corresponding heel, heel are calculated again
With the similarity between heel mirror image, the similarity as two images heel of similarity greatly is equally taken, last sole is similar
The weight of degree is 0.6, and heel weight is 0.4, and the sum of the two is as the similarity between final two images.And so on calculating
Similarity between any two width training figure.
It is further, described that discriminant function is determined according to the similarity matrix, comprising:
The similarity matrix of the training image is subjected to centralization；
The characteristic value and feature vector of similarity matrix after seeking the centralization, and give up the characteristic value less than zero pair
The feature vector answered；
According to described eigenvector computational discrimination function.
Specifically, characteristic value feature vector S corresponding less than zero will be given up, according to
Feature vector=feature vector × (1/ characteristic value)
Standardization processing is carried out, according to the eigenvectors matrix computational discrimination function after standardization, process is as follows:
A, according to the matrix H after standardization, T=H is calculatedTThe feature vector N of H-matrix.
B, the feature vector S in conjunction with features described above vector N and training image similarity matrix, which can be found out, final sentences
Other function Wherein, I is the unit matrix of m × m size, and L is m × m size and each element of matrix
It is the number of training image for 1/m, m.
Further, described to determine that the training image and image to be detected are corresponding in kernel according to the discriminant function
Training image mapping and image to be detected mapping, comprising:
According to the discriminant function training image of the same category is mapped to kernel one by one and obtains a kind of training image
Mapping；Since the coordinate that the training sample of the same category is mapped in the every dimension of kernel is all identical, therefore every a kind of formation is independent
A point.
The similarity for calculating image to be detected and all training images obtains similarity vector, and by the similarity vector
Each component sorts from large to small, and image to be detected mapping is obtained after preceding K similarity component is multiplied with the discriminant function.
It is further, described to determine whether image to be detected is new category according to the Euclidean distance, comprising:
By Euclidean distance according to sequence from small to large, and sequence first place and last N of ratio are calculated separately, by institute
The Euclidean distance value summation for stating N number of ratio and the sequence first place determines new category value；
Specifically, it is assumed that training sample classification has 5 classes, calculates separately the Euclidean distance of image to be detected Yu this five class,
Assuming that Euclidean distance and is calculated separately first place and name last, inverse according to being 1,2,3,4,5 after sorting from small to large
The ratio of second place is 0.2 and 0.25, the Euclidean distance value summation of described two ratios and the sequence first place is determined new
Class label 1.45.
By the new category value and threshold value comparison, determine that described image to be detected is if the new category value is greater than threshold value
New category determines that described image to be detected belongs to existing classification if the new category value is less than the threshold value.The present invention arranges
Flow chart is as shown in Figure 2.
The invention has the benefit that
1) present invention utilizes this nearest information of distance incessantly, it is also contemplated that the useful letter that two farthest classifications of distance include
Breath, i.e., if image to be detected belongs to known class, it is close apart from a certain classification and relatively far away from apart from other classifications；If
Image to be detected is new category, then its at a distance from training sample classification all farther out, this method can using the comparative information of distance
To increase the difference for collecting the interior outer image to be detected score of collection, Detection accuracy is improved.Image to be detected is belonging respectively in collection or collection
Outer situation is as shown in Figure 3,4 at a distance from training sample classification, and wherein triangle represents image to be detected, and other shapes represent
Training sample classification.
2) training set of the present invention when seeking discriminant function merely with known class makes criterion reach infinite, obtains
Optimal discriminant function, so that discriminant function will not be different because of the change of unknown classification sample.
3) present invention is used for the detection of new category shoe sole print, realizes the effective management printed to shoes, realizes more accurately
Shoes print is detected, is improved work efficiency.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (5)
1. a kind of shoes print new category detection method characterized by comprising
The feature of training image and image to be detected is extracted, the training image is the shoes watermark image of known class, for determining
Whether image to be detected is new category shoes watermark image；
According to similarity matrix, the training image and described image to be detected between training image described in the feature calculation
Between similarity matrix；
Discriminant function is determined according to the similarity matrix between the training image；
According to the discriminant function determine the training image and image to be detected in the mapping of kernel corresponding training image and
Image to be detected mapping；
Calculate the Euclidean distance of image to be detected mapping and training image mapping；
Determine whether image to be detected is new category according to the Euclidean distance；Include:
By Euclidean distance according to sequence from small to large, and sequence first place and last N of ratio are calculated separately, by the N
The Euclidean distance value of a ratio and the sequence first place, which is summed, determines new category value；
By the new category value and threshold value comparison, determine that described image to be detected is new class if the new category value is greater than threshold value
Not, determine that described image to be detected belongs to existing classification if the new category value is less than the threshold value.
2. the method according to claim 1, wherein the extraction training image and described image to be detected
Feature, comprising:
It will be used as sole above the training image and image to be detected according to preset value, remainder is as heel；
By the sole, heel image mirrors processing；
Sole, heel image after original image and mirror image is subjected to a quarter that wavelet transformation is the original image；
Image after the wavelet transformation is subjected to polar coordinate transform, and extracts Fourier transformation and obtains feature.
3. according to the method described in claim 2, it is characterized in that, between the training image according to the feature calculation
Similarity matrix, the similarity matrix between the training image and described image to be detected, comprising:
According to the similarity between the sole before and after training image mirror image processing described in two width of feature calculation；And compare two
Similarity, by the biggish similarity as the two width training image sole of similarity；
According to the similarity between the heel before and after training image mirror image processing described in two width of feature calculation；And compare two
Similarity, by the biggish similarity as the two width training image heel of similarity；
According to the similarity between the sole before and after the feature calculation training image and image to be detected mirror image processing；And compare
Two similarities, by the biggish similarity as sole between the training image and described image to be detected of similarity；
According to the similarity between the heel before and after the feature calculation training image and image to be detected mirror image processing；And compare
Two similarities, by the biggish similarity as heel between the training image and described image to be detected of similarity.
4. the method according to claim 1, wherein described determine discriminant function according to the similarity matrix,
Include:
The similarity matrix of the training image is subjected to centralization；
The characteristic value and feature vector of similarity matrix after seeking the centralization, and it is corresponding less than zero to give up the characteristic value
Feature vector；
According to described eigenvector computational discrimination function.
5. the method according to claim 1, wherein described determine the training image according to the discriminant function
With image to be detected in the mapping of kernel corresponding training image and image to be detected mapping, comprising:
According to the discriminant function training image of the same category is mapped to kernel one by one and obtains a kind of training image mapping；
The similarity for calculating image to be detected and all training images obtains similarity vector, and by each point of the similarity vector
Amount sorts from large to small, and is mapped in kernel after preceding K similarity component is multiplied with the discriminant function and obtains mapping to be checked
The mapping of picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610716111.1A CN106326927B (en) | 2016-08-24 | 2016-08-24 | A kind of shoes print new category detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610716111.1A CN106326927B (en) | 2016-08-24 | 2016-08-24 | A kind of shoes print new category detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106326927A CN106326927A (en) | 2017-01-11 |
| CN106326927B true CN106326927B (en) | 2019-06-04 |
Family
ID=57791986
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610716111.1A Active CN106326927B (en) | 2016-08-24 | 2016-08-24 | A kind of shoes print new category detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106326927B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778922A (en) * | 2017-02-16 | 2017-05-31 | 大连海事大学 | A kind of footwear print new category detection method of suitable high dimensional feature |
| US10621473B1 (en) * | 2019-01-30 | 2020-04-14 | StradVision, Inc. | Method for providing object detecting system capable of updating types of detectable classes in real-time by using continual learning and devices using the same |
| CN111476297A (en) * | 2020-04-07 | 2020-07-31 | 中国民航信息网络股份有限公司 | Category determination method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650944A (en) * | 2009-09-17 | 2010-02-17 | 浙江工业大学 | Method for distinguishing speakers based on protective kernel Fisher distinguishing method |
| CN105023025A (en) * | 2015-08-03 | 2015-11-04 | 大连海事大学 | Set opening trace image classification method and system |
| CN105608443A (en) * | 2016-01-22 | 2016-05-25 | 合肥工业大学 | Multi-feature description and local decision weighting face identification method |
-
2016
- 2016-08-24 CN CN201610716111.1A patent/CN106326927B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650944A (en) * | 2009-09-17 | 2010-02-17 | 浙江工业大学 | Method for distinguishing speakers based on protective kernel Fisher distinguishing method |
| CN105023025A (en) * | 2015-08-03 | 2015-11-04 | 大连海事大学 | Set opening trace image classification method and system |
| CN105608443A (en) * | 2016-01-22 | 2016-05-25 | 合肥工业大学 | Multi-feature description and local decision weighting face identification method |
Non-Patent Citations (4)
| Title |
|---|
| Kernel Null Space Methods for Novelty Detection;Bodesheim P;《IEEE Conference on Computer Vision and Pattern Recognition》;20131231;正文第3374-3381页 |
| 基于反馈的现场鞋底痕迹花纹检索算法研究;孙会会;《中国优秀硕士学位论文全文数据库》;20150315(第3期);正文第39-89页 |
| 基于相似度分布的开集人脸识别方法;张凯;《模式识别与人工智能》;20110215;第24卷(第1期);全文 |
| 核零空间线性鉴别分析及其在人脸识别中的应用;甘俊英;《计算机学报》;20141115;第37卷(第11期);全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106326927A (en) | 2017-01-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10650237B2 (en) | Recognition process of an object in a query image | |
| CN110197502B (en) | Multi-target tracking method and system based on identity re-identification | |
| CN106295124B (en) | The method of a variety of image detecting technique comprehensive analysis gene subgraph likelihood probability amounts | |
| CN103136504B (en) | Face identification method and device | |
| CN111126482B (en) | Remote sensing image automatic classification method based on multi-classifier cascade model | |
| Li et al. | Human sperm health diagnosis with principal component analysis and K-nearest neighbor algorithm | |
| CN106326927B (en) | A kind of shoes print new category detection method | |
| CN111815582A (en) | Two-dimensional code area detection method for improving background prior and foreground prior | |
| CN110874576A (en) | Pedestrian re-identification method based on canonical correlation analysis fusion features | |
| CN112613474B (en) | Pedestrian re-identification method and device | |
| Singhal et al. | Image classification using bag of visual words model with FAST and FREAK | |
| CN112307894A (en) | Pedestrian age identification method based on wrinkle features and posture features in community monitoring scene | |
| Arsa et al. | Vehicle detection using dimensionality reduction based on deep belief network for intelligent transportation system | |
| Dutra et al. | Re-identifying people based on indexing structure and manifold appearance modeling | |
| CN106778922A (en) | A kind of footwear print new category detection method of suitable high dimensional feature | |
| Bakheet et al. | Content-based image retrieval using brisk and surf as bag-of-visual-words for naïve Bayes classifier | |
| Guerrero-Peña et al. | Search-space sorting with hidden Markov models for occluded object recognition | |
| Sezavar et al. | Multi-depth deep similarity learning for person re-identification | |
| Umarani et al. | Z-Score Normalized Features with Maximum Distance Measure Based k-NN Automated Blood Cancer Diagnosis System | |
| Milgram et al. | Two-stage classification system combining model-based and discriminative approaches | |
| CN115019365B (en) | Hierarchical face recognition method based on model applicability measurement | |
| CN110390309B (en) | Finger vein illegal user identification method based on residual distribution | |
| Musa | Using Machine Learning to Overcome Facial Recognition Bias in Africa. | |
| Yuan et al. | Parameter sensitive detectors | |
| Taffar et al. | Viewpoint invariant face detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |