CN110458903A - A kind of image processing method of coded pulse sequence - Google Patents
A kind of image processing method of coded pulse sequence Download PDFInfo
- Publication number
- CN110458903A CN110458903A CN201910691729.0A CN201910691729A CN110458903A CN 110458903 A CN110458903 A CN 110458903A CN 201910691729 A CN201910691729 A CN 201910691729A CN 110458903 A CN110458903 A CN 110458903A
- Authority
- CN
- China
- Prior art keywords
- image
- information
- gray level
- pulse
- coding
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of image processing methods of coded pulse sequence, carry out multifaceted information coding to image, and the information of image different level is encoded to pulse train according to sequential relationship；It include: to convert the image into gray level image and standardization is done to the gray value of image；It is pulse by the shape of gray level image, characteristic point, color, texture coding；The pulse of coding is lined up according to setting sequence again, is combined into pulse train sequence.The present invention efficiently uses the information for the every aspect that image carries, while reducing non-critical information.For impulsive neural networks, the information completeness of the pulse train of neural network input is improved, the redundancy of information is reduced, the information processing efficiency of impulsive neural networks can be improved.For signal processing, the information capacity of information source and the code efficiency of information are improved.
Description
Technical field
The invention belongs to the calculating of class brain, impulsive neural networks, image coding technology field, and in particular to a kind of that image is compiled
Method more particularly to a kind of image processing method applied in impulsive neural networks field of information processing of the code for pulse train.
Background technique
In recent years, class brain calculating field gradually causes the concern of numerous scholars, researcher one after another biological neuron,
The fields such as connection type, impulsive neural networks expansion research, it is intended to brain be modeled to macroscopic aspect from microcosmic, further
The currently artificial intelligence processes based on depth learning technology are promoted, finally realize the target of general artificial intelligence.
Impulsive neural networks (SNN, Spiking Neuron Networks) are known as " third generation neural network ", compare
In numerical value deep learning neural network popular at present, impulsive neural networks are in connection type, information processing mechanism and dash forward
It touches and all more has imitative brain characteristic in weight learning method, therefore be also the important research direction that class brain calculates.Pulse nerve
Network processes are pulse train, and the acquisition modes of pulse train, and general there are two types of approach: one is by image according to certain
Coding mode (such as: threshold coding method, Gaussian difference point-score, frequency coding method etc.) it is converted into pulse train, another kind is direct
From neuromorphic camera.But at present due to neuromorphic camera is not mature enough in production technology, on purchasing channel not side
Just valuableness etc. and in selling price, so that the application of neuromorphic camera is not universal, there is no actually may be used by many researchers
The equipment used.Therefore, the pulse train for being now widely used in impulsive neural networks information processing is mainly derived from the first side
Formula.
Method currently used for image being encoded to pulse train is lost serious, coding inefficiency etc. there is information and is asked
Topic, and then become one of the reason for causing impulsive neural networks discrimination lower.Such as threshold coding method mentioned above,
Embodiment is one threshold value of setting, more than the pixel excitation pulse of threshold value, lower than the pixel not excitation pulse of threshold value, which
Principle is too simple so that the pulse train after coding compared to initial image, be lost a large amount of information (such as color,
Texture etc.)；Gaussian difference point-score is the local contrast of detection image, and the granting order of pulse is determined according to the power of contrast,
This method can introduce a large amount of Gaussian noise, and then can seriously affect impulsive neural networks recognition effect；Frequency coding method is to set
The frequency of fixed each pixel excitation pulse is proportional to the intensity value of the pixel, is swashed by each pixel in statistical simulation time interval
The pulse number of hair, to restore the luminous intensity of the pixel, although this mode remains the intensity signal of original image, but compile
Code piece image needs long simulation time, causes substantially reducing for recognition speed, inefficiency.
Summary of the invention
For image is encoded to during pulse in currently available technology, there is information to lose serious, code efficiency
The problems such as low, the present invention provides a kind of new pattern pulse coding method, this method extracts original image in many levels
Information greatly reduces image and is encoded into caused information loss after pulse, and encodes the pulse train fragment length of generation
Shorter (generally within 10 pulse frames), effectively increases the efficiency of coding.
Core of the invention innovative point is:
The invention proposes a kind of image processing methods that image is encoded to set of pulses sequence, to the shape of gray level image
Shape, characteristic point, color, texture are encoded；Coding is lined up according to setting sequence again, is combined into pulse train sequence.
For this method according to human eye to the view mode of object, the process of eye-observation object is not stranghtforward, incremental
: first it is seen that the appearance profile of image, then notices color, can then observe texture etc.. is substantially one
It is a from the overall situation to part again to the process of details, and the emphasis of each link concern is different.Method proposed by the present invention is to image
The information excavating of many levels is carried out, and the information of different level is encoded to pulse train according to sequential relationship.Therefore pulse
Each pulse frame of sequence features the characteristics of image of different level, when impulsive neural networks handle this pulse train,
Its treatment process is similar to the process of eye-observation things, is incremental process, is able to ascend processing result image oneself
Right property and the sense of reality.
Present invention provide the technical scheme that
A kind of pattern pulse coding method carries out multifaceted information excavating and coding to image, and by different level
Information is encoded to pulse train according to sequential relationship；The present invention comprises the steps of the method that image is encoded to pulse (such as attached
Fig. 1):
1) input picture is subjected to change of scale, is converted into the consistent image of size (long and width is calculated as H and W respectively).
2) image of same size is converted into gray level image and standardization is done to the gray value of image.
3) shape coding is done to gray level image, first using Gaussian filter then image filtering is calculated using edge detection
Method extracts the edge contour of gray level image, then profile information is done binary conversion treatment (contoured positional value is 1, no profile
Positional value be 0) after be recorded in one with gray level image long (H) and the identical two-dimensional matrix of width (W), which is counted as S1.
4) characteristic point coding is done to gray level image, first determines what key feature and the characterization key feature in image needed
Then characteristic point number extracts the position coordinates of characteristic point with feature point detection algorithm, re-define one long with gray level image
(H) it and wide (W) identical two-dimensional matrix and is initialized as 0, position identical with characteristic point coordinate in the two-dimensional matrix is arranged
It is 1, which is counted as S2；
5) color coding is done to gray level image, gray level image color (grayscale image is 256 ranks) is divided according to actual needs
It is multiple sections, defines that (H) one long and wide (W) identical as grayscale image size, high (M) three-dimensional matrice identical with division number of segment
S3, matrix initialisation 0.When dividing number of segment is 5 sections, M value is 5.
When it is implemented, by each color section respectively with the M of three-dimensional matrice S3 dimension scale be respectively 0 arrive M-1 two dimension
Matrix is corresponding；For dividing number of segment and be 5 sections, then color be divided into 0~51,52~102,103~153,154~204,
205~255 5 sections.It is respectively respectively 0,1,2,3,4 Two-Dimensional Moment with the scale tieed up of M in three-dimensional matrice S3 by each color section
Battle array is corresponding, i.e., 0~51 corresponds to S3 (M=0), and 52~102 correspond to S3 (M=1), and 103~153 correspond to S3 (M=2),
154~204 correspond to S3 (M=3), and 205~255 correspond to S3 (M=4).Then the grayscale of each pixel of gray level image is traversed
Value determines the color section i (i, value range are the integer of [0, M-1]) locating for it, and by S3 (M=i) matrix and the pixel phase
Value with position is set as 1.
6) texture coding is done to gray level image, looks first at the whole texture type of image, then extract one kind of image
Or a variety of textures, such as roughness, direction degree, the linearity etc., re-define one it is identical with gray level image long (H) and width (W)
Two-dimensional matrix is simultaneously initialized as 0, will extraction textural characteristics carry out binary conversion treatment after record in a matrix, which is counted as
S4；
7) above-mentioned matrix S1~S4 is lined up according to certain setting sequence, that is, is combined into pulse train sequence S.
The pulse train of image is obtained by the above method, while characterizing original from many levels such as profile, texture, characteristic points
Beginning image substantially increases the information utilization of original image.And the pulse train encoded with this method has
Temporal characteristics, for it is some be good at excavation, study and identify pattern pulse sequence timing information pulse neural network algorithm and
Speech (such as: the study of pattern pulse data space time information and recognition methods based on Spike cube SNN, number of patent application:
2019104814209) a kind of very matched pattern pulse sequence data type, is provided.
It should be strongly noted that handling scene, above-mentioned square for different input pictures and different pulse informations
Method corresponding to battle array S1~S4 can adjust and the sequence of S1~S4 or be made according to practical application scene
Adjustment.For some relatively simple application scenarios, can suitably reduce the information processing link of S1~S4, and for it is some compared with
For complicated application scenarios, it can according to need and suitably increase other kinds of image processing method.
Compared with prior art, the beneficial effects of the present invention are:
Using pattern pulse coding method proposed by the present invention, it is not only extraction and obtains the profile of image or to image
Light intensity is encoded, and even more carries out information excavating, each layer that effective use image carries to piece image from different levels
The information in face, and at the same time reducing some is not particularly critical information (such as light intensity of image) for identification.
For in terms of the impulsive neural networks, the present invention improves the information completeness of the pulse train of neural network input, and simultaneously
The information processing efficiency of impulsive neural networks can be improved in the redundancy for reducing information.For in terms of the signal processing, the present invention
The information capacity of information source is improved, and improves the code efficiency of information.
Detailed description of the invention
Fig. 1 is the implementation steps flow diagram of pattern pulse coding method provided by the invention.
Fig. 2 is the original image example that the embodiment of the present invention uses pulse code method to be encoded.
Fig. 3 is each dither matrix of pulse train S in the embodiment of the present invention；
Wherein, (1) is shape coding matrix S1, and (2) are characterized an encoder matrix S2, and (3)~(7) are color encoder matrix
S3, (8) are texture coding matrix S4.
Specific embodiment
With reference to the accompanying drawing, the present invention, the model of but do not limit the invention in any way are further described by embodiment
It encloses.
The present invention provides a kind of image processing method of coded pulse sequence, is a kind of new pattern pulse coding method,
This method extracts the information of original image in many levels, and the information of different level is encoded to pulse sequence according to sequential relationship
Column, greatly reduce image and are encoded into caused information loss after pulse, and encode the pulse train fragment length of generation compared with
It is short, the efficiency of coding is effectively increased, is able to solve and image is encoded to during pulse to there is information in the prior art
The problems such as losing serious, coding inefficiency.
Fig. 1 is the implementation steps flow diagram of pattern pulse coding method provided by the invention.Fig. 2 is the embodiment of the present invention
The original image example encoded using pulse code method.When it is implemented, it includes such as that image, which is encoded into pulse train,
Lower step:
1) example image is subjected to size change over, conversion growth and the wide respectively image of H and W；
2) image of size change over is converted into gray level image, and standardization is done to the gray value of image；
3) shape coding is carried out to gray level image, first using Gaussian filter to image filtering, then uses edge detection
Algorithm extracts the edge contour of gray level image, then profile information is done binary conversion treatment (contoured positional value is 1, no wheel
Wide positional value be 0) after be recorded in one with gray level image long (H) and the identical two-dimensional matrix of width (W), which is counted as
S1.(1) in Fig. 3 is the shape coding matrix S1 that the present embodiment obtains；
4) characteristic point coding is done to gray level image, first determines what key feature and the characterization key feature in image needed
Then characteristic point number extracts the position coordinates of characteristic point with feature point detection algorithm, re-define one long with gray level image
(H) it and wide (W) identical two-dimensional matrix and is initialized as 0, position identical with characteristic point coordinate in two bit matrix is arranged
It is 1, which is counted as S2；(2) in Fig. 3 are the characteristic point encoder matrix S2 that the present embodiment obtains；
5) color coding is done to gray level image, image color (grayscale image is 256 ranks) is divided into according to actual needs more
A section, defines that (H) one long and wide (W) identical as grayscale image size, high (M) three-dimensional matrice S3 identical with division number of segment, square
Battle array is initialized as 0.
When it is implemented, then color is divided into 0~51,52~102,103~153,154 for dividing number of segment and be 5 sections
~204,205~255 5 sections.Each color section is tieed up with M in three-dimensional matrice S3 to two that scale is respectively 0,1,2,3,4 respectively
It is corresponding to tie up matrix, i.e., 0~51 corresponds to S3 (M=0), and 52~102 correspond to S3 (M=1), and 103~153 correspond to S3 (M=
2), 154~204 correspond to S3 (M=3), and 205~255 correspond to S3 (M=4).Then the ash of each pixel of gray level image is traversed
Rank value determines the color section i (i, value range are the integer of [0, M-1]) locating for it, and by S3 (M=i) matrix and the pixel
The value of same position is set as 1.(3)~(7) in Fig. 3 are the color encoder matrix S3 that the present embodiment obtains.
6) texture coding is done to gray level image, looks first at the whole texture type of image, then extract one kind of image
Or a variety of textures, such as roughness, direction degree, the linearity etc., re-define one it is identical with gray level image long (H) and width (W)
Two-dimensional matrix is simultaneously initialized as 0, will extraction textural characteristics carry out binary conversion treatment after record in a matrix, which is counted as
S4；, (8) in Fig. 3 are the texture coding matrix S4 that the present embodiment obtains.
7) above-mentioned matrix S1~S4 is lined up according to certain setting sequence, that is, is combined into pulse train sequence S.
The pulse train of image is obtained by the above method, while characterizing original from many levels such as profile, texture, characteristic points
Beginning image substantially increases the information utilization of original image.And the pulse train encoded with this method has
Temporal characteristics, and for some pulse neural network algorithms for being good at excavation pulse train timing information (such as: it is based on
The pattern pulse data space time information of Spike cubeSNN learns and recognition methods, number of patent application: 2019104814209),
Provide a kind of very matched pattern pulse sequence data type.
It should be noted that the purpose for publicizing and implementing example is to help to further understand the present invention, but the skill of this field
Art personnel, which are understood that, not to be departed from the present invention and spirit and scope of the appended claims, and various substitutions and modifications are all
It is possible.Therefore, the present invention should not be limited to embodiment disclosure of that, and the scope of protection of present invention is with claim
Subject to the range that book defines.
Claims (7)
1. a kind of pattern pulse coding method, multifaceted information coding carried out to image, and by the information of image different level
Pulse train is encoded to according to sequential relationship；The following steps are included:
1) input picture is subjected to change of scale, obtains the long and wide respectively image of H and W；
2) gray level image is converted the image to, and the gray value of image is standardized；
3) shape coding is carried out to gray level image；Include:
31) first to image filtering；
32) edge contour of gray level image then is extracted, then profile information is subjected to binary conversion treatment, contoured positional value
It is 1, the positional value of no profile is 0；
33) value of binaryzation is recorded in the two-dimensional matrix of a H*W, which is counted as S1；
4) characteristic point coding is carried out to gray level image, comprising:
41) the characteristic point number that key feature and characterization key feature in image need first is determined；
42) position coordinates of characteristic point are then extracted with feature point detection algorithm；
43) it re-defines the two-dimensional matrix S2 of a H*W and is initialized as 0, it will be identical with characteristic point coordinate in the two-dimensional matrix
Position is set as 1；
5) color coding is carried out to gray level image, comprising:
51) gray level image color is 256 ranks, and gray level image color is divided into multiple sections；
52) the three-dimensional matrice S3 of a H*W*M is defined, wherein M is the number of segment that color divides；
53) three-dimensional matrice S3 is initialized as 0；
54) each color section is corresponding with the dimension M of three-dimensional matrice S3 respectively；Will each color section respectively with three-dimensional matrice
S3 M dimension scale be respectively 0 to M-1 two-dimensional matrix it is corresponding；
55) grayscale value for traversing each pixel of gray level image determines that color section i, the i value locating for it is 0~M-1；By the color
The value of the same position of the pixel is set as 1 in the dimension S3 matrix of Mi corresponding to color section；
6) texture coding is carried out to gray level image, comprising:
61) according to the texture type of image, one or more textures of image are extracted；
62) the two-dimensional matrix S4 of a H*W is defined, and is initialized as 0；
63) it is recorded in two-dimensional matrix S4 after the textural characteristics of extraction being carried out binary conversion treatment；
7) above-mentioned matrix S1~S4 is lined up according to setting sequence, that is, is combined into pulse train sequence S；
Through the above steps, realize that by image stage construction information coding be pulse train.
2. pattern pulse coding method as described in claim 1, characterized in that in step 3), specifically used Gaussian filter
To image filtering；Specifically used edge detection algorithm extracts the edge contour of gray level image.
3. pattern pulse coding method as described in claim 1, characterized in that in step 5), the number of segment that color divides is 5
Section, respectively grayscale value 0~51,52~102,103~153,154~204,205~255；Color section i value range be [0,
4] integer.
4. pattern pulse coding method as described in claim 1, characterized in that step 6) carries out texture volume to gray level image
Code, the texture includes one of roughness, direction degree, the linearity or a variety of.
5. pattern pulse coding method as described in claim 1, characterized in that according to different input pictures and different arteries and veins
Information processing scene is rushed, can adjust acquisition methods corresponding to matrix S1~S4.
6. pattern pulse coding method as described in claim 1, characterized in that matrix S1~S4 is arranged in order by step 7)
Get up to be combined into pulse train sequence S.
7. the pulse train encoded using the pattern pulse coding method described in claim 1~6 to image is answered
For the study of pattern pulse data space time information and identifying processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910691729.0A CN110458903B (en) | 2019-07-29 | 2019-07-29 | Image processing method of coding pulse sequence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910691729.0A CN110458903B (en) | 2019-07-29 | 2019-07-29 | Image processing method of coding pulse sequence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110458903A true CN110458903A (en) | 2019-11-15 |
| CN110458903B CN110458903B (en) | 2021-03-02 |
Family
ID=68483899
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910691729.0A Active CN110458903B (en) | 2019-07-29 | 2019-07-29 | Image processing method of coding pulse sequence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110458903B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112529047A (en) * | 2020-11-23 | 2021-03-19 | 广州大学 | Countermeasure sample generation method based on gradient shielding |
| CN113435455A (en) * | 2021-05-12 | 2021-09-24 | 西安工程大学 | Image contour extraction method based on space-time pulse coding |
| CN113901869A (en) * | 2021-09-02 | 2022-01-07 | 电子科技大学 | Non-contact liquid level detection method based on Spiking neural network |
| CN114222142A (en) * | 2022-02-23 | 2022-03-22 | 北京智芯微电子科技有限公司 | Image coding method and device based on impulse neural network |
| CN114466153A (en) * | 2022-04-13 | 2022-05-10 | 深圳时识科技有限公司 | Self-adaptive pulse generation method and device, brain-like chip and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194426A (en) * | 2017-05-23 | 2017-09-22 | 电子科技大学 | A kind of image-recognizing method based on Spiking neutral nets |
| US20190033407A1 (en) * | 2017-07-31 | 2019-01-31 | Samsung Electronics Co., Ltd. | Magnetic resonance imaging apparatus and method of generating magnetic resonance image |
| CN109978838A (en) * | 2019-03-08 | 2019-07-05 | 腾讯科技（深圳）有限公司 | Image-region localization method, device and Medical Image Processing equipment |
-
2019
- 2019-07-29 CN CN201910691729.0A patent/CN110458903B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194426A (en) * | 2017-05-23 | 2017-09-22 | 电子科技大学 | A kind of image-recognizing method based on Spiking neutral nets |
| US20190033407A1 (en) * | 2017-07-31 | 2019-01-31 | Samsung Electronics Co., Ltd. | Magnetic resonance imaging apparatus and method of generating magnetic resonance image |
| CN109978838A (en) * | 2019-03-08 | 2019-07-05 | 腾讯科技（深圳）有限公司 | Image-region localization method, device and Medical Image Processing equipment |
Non-Patent Citations (3)
| Title |
|---|
| DE-GAN ZHANG 等: "New Medical Image Fusion Approach with Coding Based on SCD in Wireless Sensor Network", 《JOURNAL OF ELECTRICAL ENGINEERING & TECHNOLOGY》 * |
| 王睿 等: "光学神经拟态计算研究进展", 《激光与光电子学进展》 * |
| 蔺想红 等: "基于脉冲序列核的脉冲神经元监督学习算法", 《电子学报》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112529047A (en) * | 2020-11-23 | 2021-03-19 | 广州大学 | Countermeasure sample generation method based on gradient shielding |
| CN113435455A (en) * | 2021-05-12 | 2021-09-24 | 西安工程大学 | Image contour extraction method based on space-time pulse coding |
| CN113435455B (en) * | 2021-05-12 | 2024-03-22 | 深圳灵图创新科技有限公司 | Image contour extraction method based on space-time pulse coding |
| CN113901869A (en) * | 2021-09-02 | 2022-01-07 | 电子科技大学 | Non-contact liquid level detection method based on Spiking neural network |
| CN113901869B (en) * | 2021-09-02 | 2023-04-21 | 电子科技大学 | Non-contact liquid level detection method based on Spiking neural network |
| CN114222142A (en) * | 2022-02-23 | 2022-03-22 | 北京智芯微电子科技有限公司 | Image coding method and device based on impulse neural network |
| CN114222142B (en) * | 2022-02-23 | 2022-05-06 | 北京智芯微电子科技有限公司 | Image coding method and device based on impulse neural network |
| CN114466153A (en) * | 2022-04-13 | 2022-05-10 | 深圳时识科技有限公司 | Self-adaptive pulse generation method and device, brain-like chip and electronic equipment |
| CN114466153B (en) * | 2022-04-13 | 2022-09-09 | 深圳时识科技有限公司 | Self-adaptive pulse generation method and device, brain-like chip and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110458903B (en) | 2021-03-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110458903A (en) | A kind of image processing method of coded pulse sequence | |
| CN110097131B (en) | Semi-supervised medical image segmentation method based on countermeasure cooperative training | |
| CN112052886A (en) | Human body action attitude intelligent estimation method and device based on convolutional neural network | |
| CN102592112B (en) | Method for determining gesture moving direction based on hidden Markov model | |
| CN109559287A (en) | A kind of semantic image restorative procedure generating confrontation network based on DenseNet | |
| CN109829891A (en) | A kind of magnetic tile surface defect detection method based on intensive generation confrontation neural network | |
| CN108520503A (en) | A method of based on self-encoding encoder and generating confrontation network restoration face Incomplete image | |
| CN113379601B (en) | Real world image super-resolution method and system based on degradation variable-component self-encoder | |
| CN110163815A (en) | Low-light (level) restoring method based on multistage variation self-encoding encoder | |
| CN105574929A (en) | Single vegetation three-dimensional modeling method based on ground LiDAR point cloud data | |
| CN105809715B (en) | A kind of visual movement object detection method adding up transformation matrices based on interframe | |
| CN108921019A (en) | A kind of gait recognition method based on GEI and TripletLoss-DenseNet | |
| CN110674742B (en) | Remote sensing image road extraction method based on DLinkNet | |
| CN111899353A (en) | Three-dimensional scanning point cloud hole filling method based on generation countermeasure network | |
| CN110232361B (en) | Human behavior intention identification method and system based on three-dimensional residual dense network | |
| CN110570443B (en) | Image linear target extraction method based on structural constraint condition generation model | |
| CN106650617A (en) | Pedestrian abnormity identification method based on probabilistic latent semantic analysis | |
| CN102081795B (en) | Automatic deblocking method based on sparse representation | |
| CN104899896B (en) | Multi-task learning target tracking method based on subspace characteristics | |
| CN108681689A (en) | Based on the frame per second enhancing gait recognition method and device for generating confrontation network | |
| CN100375124C (en) | A skeletonized object rebuild method | |
| CN104751111A (en) | Method and system for recognizing human action in video | |
| CN114187306A (en) | Burn area segmentation system based on impulse neural network U-shaped model | |
| CN110889868A (en) | Monocular image depth estimation method combining gradient and texture features | |
| Jiang et al. | AGD-Linknet: a road semantic segmentation model for high resolution remote sensing images integrating attention mechanism, gated decoding block and dilated convolution |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |