CN109325595A - JPEG self study quantization method and device based on traffic scene - Google Patents
JPEG self study quantization method and device based on traffic scene Download PDFInfo
- Publication number
- CN109325595A CN109325595A CN201811234868.2A CN201811234868A CN109325595A CN 109325595 A CN109325595 A CN 109325595A CN 201811234868 A CN201811234868 A CN 201811234868A CN 109325595 A CN109325595 A CN 109325595A
- Authority
- CN
- China
- Prior art keywords
- image
- data
- jpeg
- quantization
- self study
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/119—Adaptive subdivision aspects, e.g. subdivision of a picture into rectangular or non-rectangular coding blocks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/124—Quantisation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/70—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by syntax aspects related to video coding, e.g. related to compression standards
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention provides a kind of JPEG self study quantization method and device based on traffic scene, it is related to traffic video monitoring technical field, include: that image analysis is carried out to the data of input picture, obtains angular frequency data, complexity data and the frequency coefficient data of input picture;Neural network module is called to carry out the training of characteristics of image according to the angular frequency data, the complexity data and the frequency coefficient data;Real-time quantization table is calculated in the quantization parameter and Joint Photographic Experts Group quantization table for each frequency that image encoding process is exported according to the neural network module, calls self study quantization table to carry out JPEG coding, generates coded data;It decodes the coded data and generates decoding picture, the Y-PSNR of the decoding image and the input picture;Comparison result is fed back into the neural network module.The redundant data that image can be reduced reduces the physical memory space of image, greatly reduces the actual cost of traffic surveillance and control system.
Description
Technical field
The present invention relates to traffic video monitoring technical fields, certainly in particular to a kind of JPEG based on traffic scene
Study quantization method and apparatus.
Background technique
Currently, intelligent transportation is widely available, traffic block port video camera and the alert capture machine of electricity have vehicle detection, license plate
The functions such as identification, violation snap-shooting provide important leverage for maintenance traffic safety operation.Since traffic scene needs round-the-clock 24 are small
When real time monitoring capture, capture enormous amount, to network transmission and rear end storage propose high requirement.Capture the clear of image
Degree directly influences customer experience, and rear end memory space directly influences the cost of whole system.Currently, the reality of JPEG picture
Compression ratio is lower, and it is huge to capture required memory space.Therefore, how while improving candid photograph image definition, figure is reduced
The memory space of picture becomes the technical issues that need to address to reduce the actual cost of traffic surveillance and control system.
Summary of the invention
The present invention is directed at least solve one of above-mentioned technical problems existing in the prior art or related technologies, one is provided
JPEG self study quantization method and device of the kind based on traffic scene, improve monitoring image clarity, reduce image data
Memory space occupy, reduce the actual cost of traffic surveillance and control system.
The present invention is to be achieved by the following technical programs:
The first aspect of the present invention provides a kind of JPEG self study quantization method based on traffic scene, comprising: to defeated
The data for entering image carry out image analysis, obtain angular frequency data, complexity data and the frequency coefficient data of input picture,
In, described image analysis includes brightness, coloration and Analysis of Contrast, and image complexity is analyzed and DCT spectrum analysis;Call mind
The instruction of characteristics of image is carried out according to the angular frequency data, the complexity data and the frequency coefficient data through network module
Practice;Quantization parameter and Joint Photographic Experts Group the quantization meter for each frequency that image encoding process is exported according to the neural network module are calculated
Real-time quantization table is obtained, calls self study quantization table to carry out JPEG coding, generates coded data;It is raw to decode the coded data
At decoding picture, the Y-PSNR of the decoding image and the input picture;Comparison result is fed back into the mind
Through network module.
The JPEG self study quantization method based on traffic scene provided according to the present invention, it is preferable that will in cataloged procedure
The Joint Photographic Experts Group quantization table obtains the real-time quantization table multiplied by the quantization parameter.
The JPEG self study quantization method based on traffic scene provided according to the present invention, it is preferable that image analysis process
Image based on piecemeal, in image encoding process, the fritter for dividing an image into 8*8 carries out coding compression.
The JPEG self study quantization method based on traffic scene provided according to the present invention, it is preferable that image encoding process
In, according to the perception characteristics of human eye, compresses human eye and perceive insensitive image data.
The JPEG self study quantization method based on traffic scene provided according to the present invention, it is preferable that described image is complicated
Degree analysis specifically includes: carrying out image analysis using standard deviation, the complexity being calculated is as the neural network module
Input source.
The second aspect of the present invention provides a kind of JPEG self study quantization device based on traffic scene, comprising: image
Analysis module carries out image analysis for the data to input picture, obtains angular frequency data, the complexity data of input picture
And frequency coefficient data, wherein image analysis includes brightness, coloration and Analysis of Contrast, and image complexity analysis and DCT are frequently
Spectrum analysis;Training module calls neural network module to carry out figure according to angular frequency data, complexity data and frequency coefficient data
As the training of feature;Coding module, the quantization parameter for each frequency that image encoding process is exported according to neural network module and
Real-time quantization table is calculated in Joint Photographic Experts Group quantization table, calls self study quantization table to carry out JPEG coding, generates coded data;
Decoder module, decoding coded data generate decoding picture, compare the Y-PSNR of decoding image and input picture, will compare knot
Fruit feeds back to neural network module.
The JPEG self study quantization device based on traffic scene provided according to the present invention, it is preferable that will in cataloged procedure
Joint Photographic Experts Group quantifies table multiplied by quantization parameter, obtains real-time quantization table.
The JPEG self study quantization device based on traffic scene provided according to the present invention, it is preferable that image analysis process
Image based on piecemeal, in image encoding process, the fritter for dividing an image into 8*8 carries out coding compression.
The JPEG self study quantization device based on traffic scene provided according to the present invention, it is preferable that image encoding process
In, according to the perception characteristics of human eye, compresses human eye and perceive insensitive image data.
The JPEG self study quantization device based on traffic scene provided according to the present invention, it is preferable that image complexity point
Analysis specifically includes: carrying out image analysis, input source of the complexity being calculated as neural network module using standard deviation.
The beneficial effect that the present invention obtains includes at least: while guaranteeing that the image definition of human eye observation is constant, reducing
The redundant data of image reduces the physical memory space of image, greatly reduces the actual cost of traffic surveillance and control system.
Detailed description of the invention
Fig. 1 shows the realization mechanism of the JPEG self study quantization method according to an embodiment of the present invention based on traffic scene
Schematic diagram.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.
As shown in Figure 1, the course of work of the JPEG self study quantization device based on traffic scene provided according to the present invention
It specifically includes that
Image analysis is carried out to the original image of candid photograph first, image analysis module carries out brightness, Chromatic Contrast's analysis, image
The angular frequency analyzed, complexity, correction coefficient are sent into neural network module, carried out by analysis of complexity, DCT spectrum analysis
The training of characteristics of image, neural network module export the quantization parameter of each frequency, coding module by Joint Photographic Experts Group quantization table multiplied by
Quantization parameter obtains real-time quantization table, using self study quantization table carry out JPEG coding, decoder module by the image of coding into
Row decoding, PSNR (Y-PSNR) module compare original image and decoding image, and output result feeds back to neural network mould
Block is trained, and while guaranteeing picture quality, improves image compression ratio.
According to above-described embodiment, the fritter that JPEG coding divides an image into 8*8 carries out coding compression, so to image
The image that analysis is all based on piecemeal is analyzed.
Since human eye is different to the sensibility of brightness with coloration, human eye to different colours to discover threshold value also different, can be with
According to the perception characteristics of human eye, compression human eye as far as possible perceives insensitive image data, and brightness, Chromatic Contrast's analysis module are used
In the analysis of human eye characteristic.Angular frequency is a kind of visual frequency, by horizontal spatial frequency u and vertical spatial frequency v and observation
Distance d is calculated, and formula is as follows:
Image complexity carries out image analysis, input of the complexity being calculated as neural network using standard deviation
Source.
The time domain of image is converted to frequency domain by dct transform, can be used for analyzing the feature of each frequency, DCT frequency domain of the invention
Analytical procedure is as follows:
1, dct transform: compression of images is divided into 8*8 fritter, does dct transform;
2, mean value calculation: the average value of 63 ac coefficients after calculating all 8*8 sub-block dct transforms;
3, maximum value calculation: the maximum value of 63 ac coefficient average value is sought;
4, correction coefficient normalizes: the correction coefficient of 63 ac coefficients is sought by the calculated result of 2,3 steps.
The angular frequency analyzed, complexity, correction coefficient are sent into neural network module, carry out the training of characteristics of image,
Neural network module exports the quantization parameter of each frequency, gradually adjusts each layer parameter, guarantees the distortion factor and compression ratio of image.
Coding module multiplied by quantization parameter, is quantified Joint Photographic Experts Group quantization table (brightness and chromaticity quantization table) in real time
Table carries out the quantization encoding of JPEG compression using self study quantization table.
The image of coding is decoded by decoder module.
PSNR (Y-PSNR) module compares original image and decoding image, and output result feeds back to neural network mould
Block is trained, and while guaranteeing picture quality, improves image compression ratio.
According to above-described embodiment, the JPEG self study quantization method and device provided by the invention based on traffic scene is being protected
While the image definition of card human eye observation is constant, the redundant data of image is reduced, the physical memory space of image is reduced,
Greatly reduce the actual cost of traffic surveillance and control system.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of JPEG self study quantization method based on traffic scene characterized by comprising
Image analysis is carried out to the data of input picture, obtains the angular frequency data, complexity data and frequency system of input picture
Number data, wherein described image analysis includes brightness, coloration and Analysis of Contrast, and image complexity analysis and DCT frequency spectrum divide
Analysis;
Neural network module is called to carry out figure according to the angular frequency data, the complexity data and the frequency coefficient data
As the training of feature;
The quantization parameter and Joint Photographic Experts Group for each frequency that image encoding process is exported according to the neural network module quantify meter
Calculation obtains real-time quantization table, calls self study quantization table to carry out JPEG coding, generates coded data;
It decodes the coded data and generates decoding picture, the Y-PSNR of the decoding image and the input picture;
Comparison result is fed back into the neural network module.
2. the JPEG self study quantization method according to claim 1 based on traffic scene, which is characterized in that cataloged procedure
It is middle that the Joint Photographic Experts Group is quantified into table multiplied by the quantization parameter, obtain the real-time quantization table.
3. the JPEG self study quantization method according to claim 1 based on traffic scene, which is characterized in that image analysis
Image of the process based on piecemeal, in image encoding process, the fritter for dividing an image into 8*8 carries out coding compression.
4. the JPEG self study quantization method according to claim 1 based on traffic scene, which is characterized in that image coding
In the process, it according to the perception characteristics of human eye, compresses human eye and perceives insensitive image data.
5. the JPEG self study quantization method according to claim 1 based on traffic scene, which is characterized in that described image
Analysis of complexity specifically includes: carrying out image analysis using standard deviation, the complexity being calculated is as the neural network mould
The input source of block.
6. a kind of JPEG self study quantization device based on traffic scene characterized by comprising
Image analysis module carries out image analysis for the data to input picture, obtains the angular frequency data of input picture, answers
Miscellaneous degree evidence and frequency coefficient data, wherein image analysis includes brightness, coloration and Analysis of Contrast, image complexity analysis
And DCT spectrum analysis;
Training module calls neural network module according to the angular frequency data, the complexity data and the coefficient of frequency
The training of data progress characteristics of image;
Coding module, the quantization parameter and Joint Photographic Experts Group of each frequency that image encoding process is exported according to the neural network module
Real-time quantization table is calculated in quantization table, calls self study quantization table to carry out JPEG coding, generates coded data;
Decoder module decodes the coded data and generates decoding picture, the peak of the decoding image and the input picture
It is worth signal-to-noise ratio, comparison result is fed back into the neural network module.
7. the JPEG self study quantization device according to claim 6 based on traffic scene, which is characterized in that cataloged procedure
It is middle that the Joint Photographic Experts Group is quantified into table multiplied by the quantization parameter, obtain the real-time quantization table.
8. the JPEG self study quantization device according to claim 6 based on traffic scene, which is characterized in that image analysis
Image of the process based on piecemeal, in image encoding process, the fritter for dividing an image into 8*8 carries out coding compression.
9. the JPEG self study quantization device according to claim 6 based on traffic scene, which is characterized in that image coding
In the process, it according to the perception characteristics of human eye, compresses human eye and perceives insensitive image data.
10. the JPEG self study quantization device according to claim 6 based on traffic scene, which is characterized in that the figure
As analysis of complexity specifically includes: carrying out image analysis using standard deviation, the complexity being calculated is as the neural network
The input source of module.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811234868.2A CN109325595A (en) | 2018-10-23 | 2018-10-23 | JPEG self study quantization method and device based on traffic scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811234868.2A CN109325595A (en) | 2018-10-23 | 2018-10-23 | JPEG self study quantization method and device based on traffic scene |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109325595A true CN109325595A (en) | 2019-02-12 |
Family
ID=65262582
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811234868.2A Withdrawn CN109325595A (en) | 2018-10-23 | 2018-10-23 | JPEG self study quantization method and device based on traffic scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109325595A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110519595A (en) * | 2019-08-08 | 2019-11-29 | 浙江大学 | A kind of jpeg compressed image restored method based on frequency domain quantization estimated amount of damage |
CN111630570A (en) * | 2019-05-31 | 2020-09-04 | 深圳市大疆创新科技有限公司 | Image processing method, apparatus and computer-readable storage medium |
-
2018
- 2018-10-23 CN CN201811234868.2A patent/CN109325595A/en not_active Withdrawn
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111630570A (en) * | 2019-05-31 | 2020-09-04 | 深圳市大疆创新科技有限公司 | Image processing method, apparatus and computer-readable storage medium |
WO2020237646A1 (en) * | 2019-05-31 | 2020-12-03 | 深圳市大疆创新科技有限公司 | Image processing method and device, and computer-readable storage medium |
CN110519595A (en) * | 2019-08-08 | 2019-11-29 | 浙江大学 | A kind of jpeg compressed image restored method based on frequency domain quantization estimated amount of damage |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Winkler | Perceptual video quality metrics—A review | |
Thung et al. | A survey of image quality measures | |
US8743291B2 (en) | Quality assessment for images that have extended dynamic ranges or wide color gamuts | |
CN110139112B (en) | Video coding method based on JND model | |
Li et al. | A convolutional neural network-based approach to rate control in HEVC intra coding | |
Ma et al. | Reduced-reference image quality assessment in reorganized DCT domain | |
CN104219525B (en) | Perception method for video coding based on conspicuousness and minimum discernable distortion | |
Lin | Computational models for just-noticeable difference | |
Wang et al. | Quaternion representation based visual saliency for stereoscopic image quality assessment | |
CN107454389A (en) | The method for evaluating video quality and system of examining system | |
CN103051901A (en) | Video data coding device and video data encoding method | |
Wu et al. | An overview of perceptual processing for digital pictures | |
CN104322063B (en) | Method, system and the computer readable storage medium of video stream compression | |
CN112468715A (en) | Video tracing system and video tracing method | |
CN109325595A (en) | JPEG self study quantization method and device based on traffic scene | |
Wang et al. | Perceptual video coding based on saliency and just noticeable distortion for H. 265/HEVC | |
US20230362377A1 (en) | Systems, methods, and apparatuses for processing video | |
Mikhailiuk et al. | The effect of display brightness and viewing distance: a dataset for visually lossless image compression | |
Wu et al. | Perceptual VVC quantization refinement with ensemble learning | |
Zhang et al. | Visual distortion sensitivity modeling for spatially adaptive quantization in remote sensing image compression | |
CN111614962B (en) | Perceptual image compression method based on region block level JND prediction | |
CN113038129A (en) | Method and equipment for acquiring data samples for machine learning | |
CN106101486A (en) | Method of video image processing and system | |
CN112839234A (en) | Method for estimating image code rate in standard frame of screen content coding | |
Zhang et al. | Visually lossless perceptual image coding based on natural-scene masking models |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190212 |