CN110458903B - Image processing method of coding pulse sequence - Google Patents
Image processing method of coding pulse sequence Download PDFInfo
- Publication number
- CN110458903B CN110458903B CN201910691729.0A CN201910691729A CN110458903B CN 110458903 B CN110458903 B CN 110458903B CN 201910691729 A CN201910691729 A CN 201910691729A CN 110458903 B CN110458903 B CN 110458903B
- Authority
- CN
- China
- Prior art keywords
- image
- pulse
- information
- coding
- dimensional matrix
- 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
Images
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)
- Image Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses an image processing method of a coded pulse sequence, which is used for coding information of multiple layers of images and coding the information of different layers of the images into the pulse sequence according to a time sequence relation; the method comprises the following steps: converting the image into a gray image and standardizing the gray value of the image; encoding the shape, characteristic points, color and texture of the gray level image into pulses; and then arranging the coded pulses according to a set sequence to form a series of pulse sequences. The invention effectively utilizes the information of each layer carried by the image and reduces the non-key information at the same time. For the pulse neural network, the information completeness of the pulse sequence input by the neural network is improved, the redundancy of information is reduced, and the information processing efficiency of the pulse neural network can be improved. For signal processing, the information capacity of the source and the coding efficiency of the information are improved.
Description
Technical Field
The invention belongs to the technical field of brain-like computation, impulse neural network and image coding, in particular relates to a method for coding an image into an impulse sequence, and particularly relates to an image processing method applied to the field of impulse neural network information processing.
Background
In recent years, the field of brain-like computing has gradually attracted the attention of numerous scholars, researchers have developed researches in the fields of biological neurons, connection modes, impulse neural networks and the like, and try to model the brain from a microscopic level to a macroscopic level, further promote the current artificial intelligence process based on a deep learning technology, and finally achieve the aim of general artificial intelligence.
Compared with the widely popular numerical deep learning neural network, the Spiking neural network has more brain-like characteristics in connection mode, information processing mechanism and synaptic weight learning method, and is an important research direction of brain-like calculation. The impulse neural network processes the impulse sequence, and the acquisition mode of the impulse sequence generally has two approaches: one is to convert the image into a pulse sequence according to a certain coding mode (such as a threshold coding method, a Gaussian difference method, a frequency coding method and the like), and the other is to directly come from a neuromorphic camera. However, the neuromorphic camera is not widely used and many researchers do not have a practical device because the neuromorphic camera is not mature enough in production process, inconvenient in purchase channel, expensive in sale price and the like. Therefore, the pulse sequence widely used for information processing of the impulse neural network at present mainly originates from the first mode.
The current method for coding the image into the pulse sequence has the problems of serious information loss, low coding efficiency and the like, and further becomes one of the reasons of low recognition rate of the pulse neural network. For example, the threshold encoding method mentioned above is implemented by setting a threshold value, the pixels above the threshold value excite the pulse, and the pixels below the threshold value do not excite the pulse, and the principle of the method is too simple, so that the encoded pulse sequence loses a large amount of information (such as color, texture, etc.) compared with the original image; the Gaussian difference method is used for detecting the local contrast of an image and determining the issuing sequence of pulses according to the strength of the contrast, and a large amount of Gaussian noise is introduced by the method so as to seriously influence the identification effect of the pulse neural network; the frequency coding method is to set the frequency of each pixel excitation pulse to be proportional to the intensity value of the pixel, and restore the light intensity of the pixel by counting the number of pulses excited by each pixel in the simulation time interval.
Disclosure of Invention
Aiming at the problems of serious information loss, low coding efficiency and the like in the process of coding an image into pulses in the prior art, the invention provides a novel image pulse coding method, which extracts the information of an original image on a plurality of layers, greatly reduces the information loss caused after the image is coded into pulses, and effectively improves the coding efficiency because the length of a pulse sequence segment generated by coding is shorter (generally within 10 pulse frames).
The core innovation points of the invention are as follows:
the invention provides an image processing method for encoding an image into a group of pulse sequences, which encodes the shape, characteristic points, color and texture of a gray level image; and then the codes are arranged according to a set sequence to form a series of pulse sequences. According to the method, the process of observing the object by the human eyes is not kicking but progressive: looking first at the outline of the image, then noting the colors, then it will be observed that texture etc. … … is essentially a process from global to local to detail, with each step focusing on a different emphasis. The method provided by the invention carries out information mining on a plurality of layers of images and encodes information of different layers into pulse sequences according to a time sequence relation. Therefore, each pulse frame of the pulse sequence describes image characteristics of different layers, when the pulse neural network processes the pulse sequence, the processing process is similar to the process of observing objects by human eyes and is a progressive process, and the naturalness and the reality of an image processing result can be improved.
The technical scheme provided by the invention is as follows:
a image pulse coding method, carry on the information mining and coding of the multi-layer to the picture, and encode the information of different layers into the pulse sequence according to the time sequence relation; the method for encoding the image into the pulse comprises the following steps (as shown in the attached figure 1):
1) the input image is subjected to scale conversion and converted into images of uniform size (length and width are denoted as H and W, respectively).
2) And converting the image with the same size into a gray image and standardizing the gray value of the image.
3) The gray image is subjected to shape coding, a Gaussian filter is used for filtering the image, then an edge detection algorithm is used for extracting an edge contour of the gray image, contour information is subjected to binarization processing (the position value with the contour is 1, and the position value without the contour is 0) and then recorded in a two-dimensional matrix with the same length (H) and width (W) as the gray image, and the matrix is counted as S1.
4) Performing feature point coding on the gray level image, firstly determining key features in the image and the number of feature points required by the representation of the key features, then extracting position coordinates of the feature points by using a feature point detection algorithm, then defining a two-dimensional matrix with the same length (H) and width (W) as the gray level image and initializing the two-dimensional matrix to be 0, setting the position of the two-dimensional matrix with the same coordinate as the feature points to be 1, and counting the matrix as S2;
5) the gray image is color coded, the color of the gray image (the gray image is 256 steps) is divided into a plurality of segments according to actual needs, a three-dimensional matrix S3 with the length (H) and the width (W) being the same as the size of the gray image and the height (M) being the same as the number of the divided segments is defined, and the matrix is initialized to be 0. When the number of the divided sections is 5, the value of M is 5.
In specific implementation, each color segment corresponds to a two-dimensional matrix with the scale of M dimension of the three-dimensional matrix S3 from 0 to M-1; for example, if the number of segments is 5, the colors are divided into five segments, i.e., 0 to 51, 52 to 102, 103 to 153, 154 to 204, and 205 to 255. Each color segment is respectively corresponding to a two-dimensional matrix with the scale of the mth dimension of 0, 1, 2, 3 and 4 in the three-dimensional matrix S3, namely 0 to 51 correspond to S3(M is 0), 52 to 102 correspond to S3(M is 1), 103 to 153 correspond to S3(M is 2), 154 to 204 correspond to S3(M is 3), and 205 to 255 correspond to S3(M is 4). Then, the gray-scale value of each pixel of the gray-scale image is traversed, the color segment i (i, an integer with a value range of [0, M-1 ]) where the pixel is located is determined, and the value of the same position of the S3(M ═ i) matrix and the pixel is set to be 1.
6) Performing texture coding on a gray level image, firstly observing the integral texture type of the image, then extracting one or more textures of the image, such as roughness, direction degree, linearity and the like, then defining a two-dimensional matrix with the length (H) and width (W) same as the gray level image and initializing the two-dimensional matrix to 0, performing binarization processing on the extracted texture characteristics and recording the texture characteristics in the matrix, wherein the matrix is counted as S4;
7) the matrices S1 to S4 are arranged in a predetermined order and combined into a series of pulse trains S.
The pulse sequence of the image is obtained by the method, and the original image is represented from a plurality of layers such as the contour, the texture, the feature points and the like, so that the information utilization rate of the original image is greatly improved. The pulse sequence coded by the method has time sequence characteristics, and provides a very matched image pulse sequence data type for some pulse neural network algorithms which are good at mining, learning and identifying image pulse sequence time sequence information (for example, a Spike cube SNN-based image pulse data space-time information learning and identifying method, patent application number: 2019104814209).
It should be noted that, for different input images and different pulse information processing scenarios, the method corresponding to the matrixes S1-S4 can be adjusted, and the sequence of S1-S4 can also be adjusted according to the actual application scenario. For some simpler application scenes, the information processing links from S1 to S4 can be reduced appropriately, and for some more complex application scenes, other types of image processing methods can be added appropriately according to needs.
Compared with the prior art, the invention has the beneficial effects that:
the image pulse coding method provided by the invention not only extracts the outline of the image or codes the light intensity of the image, but also carries out information mining on the image from different layers, effectively utilizes the information of each layer carried by the image, and simultaneously reduces some information (such as the light intensity of the image) which is not particularly critical to recognition. In terms of the pulse neural network, the invention improves the information completeness of the pulse sequence input by the neural network, reduces the redundancy of information and can improve the information processing efficiency of the pulse neural network. From the aspect of signal processing, the invention improves the information capacity of the information source and improves the coding efficiency of the information.
Drawings
Fig. 1 is a flow chart illustrating an implementation procedure of a method for encoding image pulses according to the present invention.
Fig. 2 is an example of an original image encoded by a pulse encoding method according to an embodiment of the present invention.
FIG. 3 is a diagram of each pulse matrix of the pulse sequence S according to an embodiment of the present invention;
the shape coding matrix S1 (1), the feature point coding matrix S2 (2), the color coding matrix S3 (3) to (7), and the texture coding matrix S4 (8) are listed below.
Detailed Description
The invention will be further described by way of examples, without in any way limiting the scope of the invention, with reference to the accompanying drawings.
The invention provides an image processing method of a coding pulse sequence, which is a new image pulse coding method, and the method extracts the information of an original image from a plurality of layers and codes the information of different layers into the pulse sequence according to the time sequence relation, thereby greatly reducing the information loss caused after the image is coded into the pulse, and the length of the pulse sequence segment generated by coding is shorter, thereby effectively improving the coding efficiency, and solving the problems of serious information loss, low coding efficiency and the like in the process of coding the image into the pulse in the prior art.
Fig. 1 is a flow chart illustrating an implementation procedure of a method for encoding image pulses according to the present invention. Fig. 2 is an example of an original image encoded by a pulse encoding method according to an embodiment of the present invention. In specific implementation, encoding an image into a pulse sequence comprises the following steps:
1) transforming the example image into an image with a length and a width of H and W respectively by size transformation;
2) converting the size-converted image into a gray image, and standardizing the gray value of the image;
3) the gray image is subjected to shape coding, a Gaussian filter is used for filtering the image, then an edge detection algorithm is used for extracting an edge contour of the gray image, contour information is subjected to binarization processing (the position value with the contour is 1, and the position value without the contour is 0) and then recorded in a two-dimensional matrix with the same length (H) and width (W) as the gray image, and the matrix is counted as S1. Fig. 3 shows (1) that is the shape coding matrix S1 obtained in this embodiment;
4) performing feature point coding on the gray level image, firstly determining key features in the image and the number of feature points required by the representation of the key features, then extracting position coordinates of the feature points by using a feature point detection algorithm, then defining a two-dimensional matrix with the same length (H) and width (W) as the gray level image and initializing the two-dimensional matrix to be 0, setting the position of the two-dimensional matrix with the same coordinate as the feature points to be 1, and counting the matrix as S2; fig. 3 (2) is the characteristic point coding matrix S2 obtained in this embodiment;
5) the gray image is color coded, the image color (the gray image is 256 steps) is divided into a plurality of segments according to the actual requirement, a three-dimensional matrix S3 with the length (H) and the width (W) being the same as the size of the gray image and the height (M) being the same as the number of the divided segments is defined, and the matrix is initialized to be 0.
In the specific implementation, for example, the number of the divided segments is 5, the colors are divided into five segments of 0 to 51, 52 to 102, 103 to 153, 154 to 204, and 205 to 255. Each color segment is respectively corresponding to a two-dimensional matrix with the M-th dimension scale of 0, 1, 2, 3 and 4 in the three-dimensional matrix S3, namely 0-51 corresponds to S3(M is 0), 52-102 corresponds to S3(M is 1), 103-153 corresponds to S3(M is 2), 154-204 corresponds to S3(M is 3), and 205-255 corresponds to S3(M is 4). Then, the gray-scale value of each pixel of the gray-scale image is traversed, the color segment i (i, an integer with a value range of [0, M-1 ]) where the pixel is located is determined, and the value of the same position of the S3(M ═ i) matrix and the pixel is set to be 1. The color coding matrix S3 obtained in the present embodiment is (3) to (7) in fig. 3.
6) Performing texture coding on a gray level image, firstly observing the integral texture type of the image, then extracting one or more textures of the image, such as roughness, direction degree, linearity and the like, then defining a two-dimensional matrix with the length (H) and width (W) same as the gray level image and initializing the two-dimensional matrix to 0, performing binarization processing on the extracted texture characteristics and recording the texture characteristics in the matrix, wherein the matrix is counted as S4; fig. 3 (8) shows the texture coding matrix S4 obtained in this embodiment.
7) The matrices S1 to S4 are arranged in a predetermined order and combined into a series of pulse trains S.
The pulse sequence of the image is obtained by the method, and the original image is represented from a plurality of layers such as the contour, the texture, the feature points and the like, so that the information utilization rate of the original image is greatly improved. The pulse sequence coded by the method has time sequence characteristics, and a very matched image pulse sequence data type is provided for some pulse neural network algorithms which are good at mining the time sequence information of the pulse sequence (for example, a Spike cube SNN-based image pulse data space-time information learning and identifying method, patent application number: 2019104814209).
It is noted that the disclosed embodiments are intended to aid in further understanding of the invention, but those skilled in the art will appreciate that: various substitutions and modifications are possible without departing from the spirit and scope of the invention and appended claims. Therefore, the invention should not be limited to the embodiments disclosed, but the scope of the invention is defined by the appended claims.
Claims (6)
1. A image pulse coding method, carry on the information coding of the multi-layer to the picture, and encode the information of different layers of picture as the pulse sequence according to the time sequence relation; the method comprises the following steps:
1) carrying out scale transformation on an input image to obtain images with the length and width of H and W respectively;
2) converting the image into a gray level image, and carrying out standardization processing on the gray level value of the image;
3) carrying out shape coding on the gray level image; the method comprises the following steps:
31) filtering an image;
32) then extracting the edge contour of the gray level image, and then carrying out binarization processing on contour information, wherein the position value with the contour is 1, and the position value without the contour is 0;
33) recording the binarized values in a two-dimensional matrix of H x W, which is designated S1;
4) carrying out feature point coding on the gray level image, comprising the following steps:
41) firstly, determining key features in an image and the number of feature points required for representing the key features;
42) then extracting the position coordinates of the feature points by using a feature point detection algorithm;
43) defining a two-dimensional matrix S2 of H x W, initializing the two-dimensional matrix to 0, and setting the position in the two-dimensional matrix, which is the same as the coordinate of the characteristic point, as 1;
5) color coding a grayscale image, comprising:
51) the gray level image color is 256 steps, and the gray level image color is divided into a plurality of segments;
52) defining a three-dimensional matrix S3 of H W M, where M is the number of segments of the color division;
53) initializing the three-dimensional matrix S3 to 0;
54) each color segment is respectively corresponding to the dimension M of the three-dimensional matrix S3; namely, each color segment corresponds to a two-dimensional matrix with the scale of M dimension of the three-dimensional matrix S3 from 0 to M-1;
55) traversing the gray level value of each pixel of the gray level image, and determining the color segment i where the pixel is located, wherein the value of i is 0-M-1; setting the value of the same position of the pixel in an S3 matrix of dimension M-i corresponding to the color segment as 1;
6) texture coding a grayscale image, comprising:
61) extracting one or more textures of the image according to the texture type of the image;
62) defining a two-dimensional matrix S4 of H W and initializing to 0;
63) performing binarization processing on the extracted texture features and recording the processed texture features in a two-dimensional matrix S4;
7) arranging the matrixes S1-S4 according to a set sequence to form a series of pulse sequences S;
through the steps, the multi-layer information of the image is coded into the pulse sequence.
2. The method for pulse encoding of an image as claimed in claim 1, wherein in step 3), the image is filtered, in particular using a gaussian filter; specifically, an edge detection algorithm is used for extracting the edge contour of the gray level image.
3. The method for encoding an image pulse according to claim 1, wherein in the step 5), the number of the color division is 5, and the gray scale values are 0 to 51, 52 to 102, 103 to 153, 154 to 204, and 205 to 255; the color segment i is an integer with the value range of [0,4 ].
4. The image pulse encoding method of claim 1, wherein step 6) texture-encodes the gray scale image, said texture including one or more of roughness, orientation, and linearity.
5. The image pulse encoding method of claim 1, wherein the acquisition methods corresponding to the matrices S1 to S4 are adjustable according to different input images and different pulse information processing scenes.
6. A method as claimed in any one of claims 1 to 5, wherein the pulse sequence is applied to image pulse data spatio-temporal information learning and identification processes.
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 CN110458903A (en) | 2019-11-15 |
| CN110458903B true 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) |
Families Citing this family (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 |
| 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 |
| 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 |
| 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 |
| 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 |
Citations (2)
| 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 |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109978838B (en) * | 2019-03-08 | 2021-11-30 | 腾讯科技（深圳）有限公司 | Image area positioning method and device and medical image processing equipment |
-
2019
- 2019-07-29 CN CN201910691729.0A patent/CN110458903B/en active Active
Patent Citations (2)
| 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 |
Non-Patent Citations (3)
| Title |
|---|
| New Medical Image Fusion Approach with Coding Based on SCD in Wireless Sensor Network;De-gan Zhang 等;《Journal of Electrical Engineering & Technology》;20151231;第6卷(第10期);第2384-2392页 * |
| 光学神经拟态计算研究进展;王睿 等;《激光与光电子学进展》;20160921;第120004-1-120004-11页 * |
| 基于脉冲序列核的脉冲神经元监督学习算法;蔺想红 等;《电子学报》;20161215;第44卷(第12期);第2877-2886页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110458903A (en) | 2019-11-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110458903B (en) | Image processing method of coding pulse sequence | |
| CN110084156B (en) | Gait feature extraction method and pedestrian identity recognition method based on gait features | |
| CN109615582B (en) | Face image super-resolution reconstruction method for generating countermeasure network based on attribute description | |
| CN107967695B (en) | A kind of moving target detecting method based on depth light stream and morphological method | |
| CN110458844B (en) | Semantic segmentation method for low-illumination scene | |
| CN104992223B (en) | Intensive Population size estimation method based on deep learning | |
| CN110555368A (en) | Fall-down behavior identification method based on three-dimensional convolutional neural network | |
| CN108133188A (en) | A kind of Activity recognition method based on motion history image and convolutional neural networks | |
| CN111832516B (en) | Video behavior recognition method based on unsupervised video representation learning | |
| CN109993269B (en) | Single image crowd counting method based on attention mechanism | |
| CN111104903A (en) | Depth perception traffic scene multi-target detection method and system | |
| CN106960176B (en) | Pedestrian gender identification method based on transfinite learning machine and color feature fusion | |
| CN106650617A (en) | Pedestrian abnormity identification method based on probabilistic latent semantic analysis | |
| CN105787867A (en) | Method and apparatus for processing video images based on neural network algorithm | |
| CN115619743A (en) | Construction method and application of OLED novel display device surface defect detection model | |
| CN108681689A (en) | Based on the frame per second enhancing gait recognition method and device for generating confrontation network | |
| Bounsaythip et al. | Genetic algorithms in image processing-a review | |
| CN112766062A (en) | Human behavior identification method based on double-current deep neural network | |
| CN1858773A (en) | Image identifying method based on Gabor phase mode | |
| CN110889868A (en) | Monocular image depth estimation method combining gradient and texture features | |
| CN112184731B (en) | Multi-view stereoscopic depth estimation method based on contrast training | |
| CN113537240B (en) | Deformation zone intelligent extraction method and system based on radar sequence image | |
| CN114463614A (en) | Significance target detection method using hierarchical significance modeling of generative parameters | |
| CN103971385B (en) | The detection method of moving objects in video | |
| CN110188706B (en) | Neural network training method and detection method based on character expression in video for generating confrontation network |
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 |