CN104661023A - Image or video coding method based on predistortion and training filter - Google Patents
Image or video coding method based on predistortion and training filter Download PDFInfo
- Publication number
- CN104661023A CN104661023A CN201510058184.1A CN201510058184A CN104661023A CN 104661023 A CN104661023 A CN 104661023A CN 201510058184 A CN201510058184 A CN 201510058184A CN 104661023 A CN104661023 A CN 104661023A
- Authority
- CN
- China
- Prior art keywords
- image
- filter
- training
- predistortion
- video
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention relates to an image or video compression coding and provides an image or video coding method based on predistortion and filter training, and the method not only can improve the coding efficiency but also avoids a ringing effect so as to guarantee the image quality. The image or video coding method based on predistortion and filter training comprises the following steps: through a low-pass filter, performing predistortion fuzzification treatment on a sequence of each frame of video before coding; at a decoding end, adopting a large amount of images with various image structures to form a training sample library at first; then, performing offline training on a filter set by the classified images; when reconstructing fuzzy images at the decoding end, classifying the image structures by adopting a combined method of ADRC (adaptive dynamic range coding) and the other classification method, finding corresponding optimal filter coefficients in a lookup table according to a classification result, forming optimal filters, performing defuzzification respectively based on image structure types, and finally performing synthesis to obtain defuzzified images. The image or video coding method is mainly applied to image or video compression coding.
Description
Technical field
The present invention relates to image, video compression coding, particularly based on predistortion and image or the method for video coding of training filter.
Background technology
Along with the development of electronic equipment, the Internet and improving constantly of people's demand, digital picture and video data produce just at a terrific speed and propagate, and high definition or ultra high-definition video become development trend.Such as, 3D, HDR (high dynamic range), 4K or 8K video etc., storing or transmitting these video datas all needs efficient compression method to reduce data volume and to ensure the quality of image.
Major video compression standard mainly contain MPEG-2, MPEG-4, H.264/AVC, HEVC etc.These compression and coding standards utilize video sequence spatial coherence and temporal correlation, remove the bulk redundancy information existed in video sequence, only retain a small amount of uncorrelated information and transmit, to reduce code check, to save transmission bandwidth.And receiver utilizes these irrelevant information, according to certain decoding algorithm, original video sequence content can be recovered under the prerequisite ensureing certain picture quality.Estimation in video compression coding and dct transform are the methods of important minimizing signal redundancy.Estimation make use of the local directivity architectural characteristic of image, and DCT make use of the grading structure characteristic that natural image has, i.e. low-pass characteristic, is intended to remove the redundancy in block between pixel.Dct transform by concentration of energy on a small amount of low frequency coefficient, can more easily realize compression in conjunction with quantification and entropy code.These methods all contribute to the minimizing of amount of coded data, and in view of current multi-medium data amount is increasing, under limited storage and bandwidth resources condition, improving video compression coding efficiency further has important realistic meaning.
2007, Lei.Zhichun [1] proposed a kind of video compression coding scheme, and this scheme at coding side predistortion Fuzzy processing video, and transmits ambiguity function in code stream, when decoding end reconstructs by this ambiguity function deconvolution deblurring.The method effectively can reduce video data to be encoded amount and then reduce video stream code rate, but because using deconvolution deblurring to produce ringing effect when reconstructing.
Ringing effect is the inherent shortcoming of deconvolution deblurring method.In image processing field, researcher is devoted to find a kind of new deblurring algorithm to replace deconvolution always.1998, the people such as T.Kondo [2] proposed a kind of least-mean-square filter based on Images Classification, had good effect for image enhaucament.Adaptive dynamic range coding (ADRC) algorithm that picture structure of giving chapter and verse again afterwards carries out classifying is to recover original image [3].2008, the people [4] such as Ling Shao propose the algorithm Recovery image that ADRC combines with another kind of image classification method, and test for the algorithm that ADRC and mean absolute difference (MAD), ADRC and standard deviation (STD), ADRC and dynamic range (DR) combine, effect is better than cascading filter.2014, nightstool Yang Dengren [5] proposes a kind of boundary effect Restrainable algorithms, utilize the convolution pyramid filter group model with symmetry coefficient, one group of filter coefficient is trained in the region of type every in certain particular image, and train the bank of filters obtained to be applied to the continuation region solving other image respective type this, the method computational speed is fast, effectively can suppress the ringing effect of various Frequency Domain Deconvolution algorithm.
In sum, existing research proves that coding side predistortion Fuzzy Processing can reduce code check, training filtered method can realize image deblurring and avoid ringing effect (only for image procossing, not relating to coding) respectively.Not yet find that there is these two kinds of methods combining to get up for image and video coding field at present.
Therefore, the present invention proposes a kind of based on predistortion and the training image of filter or Video Coding Scheme, namely uses predistortion (obfuscation) process to reduce code check, adopt off-line training is good filter to carry out the compression coding scheme of deblurring process Recovery image quality in decoding end at coding side.
List of references
[1] L.Zhichun.Signal coding and decoding with pre-and post-processing. Europe, 06006924.2 [P], 2007-10-03.
[2]T.Kondo and K.Kawaguchi,Adaptive dynamic range encoding method andapparatus.U.S.patent 5444487,Aug.1998.
[3]T.Kondo,Y.Node,T.Fujiwara,and Y.Okumura,Picture conversion apparatus.pictureconversion method,learning apparatus and learning method,U.S.patent 6323905,Nov.2001.
[4]Ling Shao,Hui Zhang,and Gerard de Haan.An Overview and Performance Evaluation ofClassification-Base Least Squares Trained Filters[C].//Transaction on ImageProcessing,2008,17(10):1772-1782.
[5] nightstool is raised, Liu Xuehui, Wu Enhua. the quick Restrainable algorithms [J] of image deconvolution boundary effect. and computer-aided design and graphics journal, 2014,26 (7): 1051-1066.
Summary of the invention
For overcoming the deficiencies in the prior art, providing one to improve code efficiency, avoiding ringing effect again, thus ensure picture quality based on predistortion and the training image of filter or method for video coding.For this reason, the technical scheme that the present invention takes is, based on predistortion and image or the method for video coding of training filter, by low pass filter, first carries out predistortion Fuzzy processing to every frame video sequence before the coding;
In decoding end, first adopt the image composition training sample database in a large number with various picture structure, the difference according to picture structure is classified to zones of different, and the concrete method adopting ADRC to combine with another kind of sorting technique is classified to image;
Then, carry out off-line training with sorted image to bank of filters, adopt the training algorithm of least mean-square error, the filter coefficient of one group of optimum is trained in the region for every type, forms look-up table (LUT), stored in decoding end reconstructed module;
After decoding end reconstructs blurred picture, the method adopting ADRC to combine with another sorting technique is classified to picture structure, corresponding optimal filter coefficients composition optimal filter is found in a lookup table according to classification results, by picture structure classification deblurring respectively, finally synthesis obtains the image after deblurring.
Picture structure specifically comprises smooth region, texture region, fringe region.
Another kind of sorting technique is specially the one in local entropy, mean absolute difference (MAD), standard deviation (STD), dynamic range (DR).
When decoded reconstructed image, if run into the picture structure do not had in training sample database, namely the filter coefficient of this structure corresponding is not had in look-up table, Deconvolution Method is then adopted to replace training filter Recovery image or video, by Deconvolution Method as a kind of alternative scheme outside training filter deblurring algorithm, for recovering original image or video.
Compared with the prior art, technical characterstic of the present invention and effect:
The present invention proposes based on predistortion and image or the Video Coding Scheme of training filter, and the predistortion Fuzzy Processing of coding side can reduce code check, improve code efficiency.The training filter deblurring algorithm of decoding and reconstituting end selects different filters to carry out deblurring recovery original image according to the difference of picture structure, can avoid ringing effect, improve decoded picture or video quality.
Accompanying drawing explanation
Fig. 1 the present invention propose based on predistortion and the training image of filter or Video Coding Scheme.
Embodiment
Multi-medium data amount is increasing, and consumer requires more and more higher to multimedia quality simultaneously.Therefore need to study and can improve code efficiency further and the algorithm ensureing certain coding quality.
Carry out predistortion Fuzzy processing at coding side can reduce data to be encoded amount, improve code efficiency, but challenge is proposed to the deblurring function of decoding end reconstructed module.If transmission ambiguity function can increase code stream expense, and traditional deconvolution image deblurring algorithm can cause ringing effect.Therefore, the predistortion-deblurring scheme being applicable to Video coding is developed significant.
The present invention proposes a kind of based on predistortion and image or the Video Coding Scheme of training filter.This scheme carries out predistortion (obfuscation) process at coding side to image or video, and without the need to transfer encoding end predistortion (obfuscation) function in code stream, select the good different filter of off-line training to carry out deblurring in decoding end according to different images structure.Can either code efficiency be improved, avoid ringing effect again, ensure that picture quality.
In existing encoding scheme, usually, will carry out low-pass filtering before the coding to remove high-frequency noise, the mode of this low-pass filtering does not have a significant effect to picture quality.For reducing data to be encoded amount further, the present invention first carries out predistortion Fuzzy processing to every frame video sequence before the coding, it can be made to be realized by the low pass filter that size is 5 × 5, and then carry out compressed encoding.The fuzzy high fdrequency component of image that makes significantly reduces, and compression efficiency is higher, but Fuzzy processing makes image fault, needs to adopt deblurring algorithm Recovery image quality when decoding end reconstructs.
The algorithm of training filter deblurring is used for decoding end reconstructed module by the present invention.First adopt and there is various picture structure (smooth region in a large number, texture region, fringe region etc.) image composition training sample database, difference according to picture structure is classified to zones of different, usually, the different structure of method to image of adaptive dynamic range coding (ADRC) can be adopted to classify, but the method does not have the difference of differentiate between images high-contrast and low contrast structure and does not consider the problem that object detail Partial shrinkage noise is identical, the method that ADRC therefore can be adopted to combine with another kind of sorting technique is classified to image.Such as, the method etc. that combines of ADRC and local entropy, ADRC and mean absolute difference (MAD), ADRC and standard deviation (STD), ADRC and dynamic range (DR).
Then, carry out off-line training with sorted image to bank of filters, can adopt the training algorithm of least mean-square error, the filter coefficient of one group of optimum is trained in the region for every type, form look-up table (LUT), stored in decoding end reconstructed module.
After decoding end reconstructs blurred picture, the method adopting ADRC to combine with another sorting technique (as local entropy) is classified to picture structure, corresponding optimal filter coefficients composition optimal filter is found in a lookup table according to classification results, by picture structure classification deblurring respectively, finally synthesis obtains the image after deblurring.
In addition, when decoded reconstructed image, if run into the picture structure (namely not having the filter coefficient of this structure corresponding in look-up table) do not had in training sample database, Deconvolution Method can be adopted to replace training filter Recovery image or video, namely Deconvolution Method can as a kind of alternative scheme outside training filter deblurring algorithm, for recovering original image or video.
The present invention proposes a kind of based on predistortion and image or the Video Coding Scheme of training filter.Wherein, in filter training, the method adopting adaptive dynamic range coding (ADRC) and local entropy (or ADRC and mean absolute difference (MAD), ADRC and standard deviation (STD), ADRC and dynamic range (DR)) to combine is classified for the different structure of image, adopt the training algorithm of least mean-square error afterwards, one group of filter coefficient is trained in region for every type, form look-up table (LUT), stored in decoding end reconstructed module.In the implementation process of whole scheme, first, at coding side, pre-distortion is carried out to reduce code check to input picture or video sequence; Then according to existing encoding scheme, encoding and decoding are carried out to the image after predistortion or video; Finally, when decoded reconstructed image, using the filter that off-line training completes, according to searching best filter factor in the look-up table that do not coexist (LUT) of picture structure, completing the reconstruction of image or video.
Claims (4)
1., based on predistortion and image or a method for video coding of training filter, it is characterized in that, comprise the steps:
At coding side: by low pass filter, first predistortion Fuzzy processing is carried out to every frame video sequence before the coding;
In decoding end, first adopt the image composition training sample database in a large number with various picture structure, the difference according to picture structure is classified to zones of different, and the concrete method adopting ADRC to combine with another kind of sorting technique is classified to image;
Then, carry out off-line training with sorted image to bank of filters, adopt the training algorithm of least mean-square error, the filter coefficient of one group of optimum is trained in the region for every type, forms look-up table (LUT), stored in decoding end reconstructed module;
After decoding end reconstructs blurred picture, the method adopting ADRC to combine with another sorting technique is classified to picture structure, corresponding optimal filter coefficients composition optimal filter is found in a lookup table according to classification results, by picture structure classification deblurring respectively, finally synthesis obtains the image after deblurring.
2. according to claim 1 based on predistortion and image or the method for video coding of training filter, it is characterized in that, picture structure specifically comprises smooth region, texture region, fringe region.
3. according to claim 1 based on predistortion and image or the method for video coding of training filter, it is characterized in that, another kind of sorting technique is specially the one in local entropy, mean absolute difference (MAD), standard deviation (STD), dynamic range (DR).
4. according to claim 1 based on predistortion and image or the method for video coding of training filter, it is characterized in that, when decoded reconstructed image, if run into the picture structure do not had in training sample database, namely the filter coefficient of this structure corresponding is not had in look-up table, Deconvolution Method is then adopted to replace training filter Recovery image or video, by Deconvolution Method as a kind of alternative scheme outside training filter deblurring algorithm, for recovering original image or video.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510058184.1A CN104661023B (en) | 2015-02-04 | 2015-02-04 | Image or method for video coding based on predistortion and training wave filter |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510058184.1A CN104661023B (en) | 2015-02-04 | 2015-02-04 | Image or method for video coding based on predistortion and training wave filter |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104661023A true CN104661023A (en) | 2015-05-27 |
| CN104661023B CN104661023B (en) | 2018-03-09 |
Family
ID=53251651
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510058184.1A Expired - Fee Related CN104661023B (en) | 2015-02-04 | 2015-02-04 | Image or method for video coding based on predistortion and training wave filter |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104661023B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110149554A (en) * | 2019-05-31 | 2019-08-20 | Oppo广东移动通信有限公司 | Method, apparatus, electronic equipment and the storage medium of video image processing |
| CN110598584A (en) * | 2019-08-26 | 2019-12-20 | 天津大学 | Convolutional neural network face recognition algorithm based on wavelet transform and DCT |
| CN112805990A (en) * | 2018-11-15 | 2021-05-14 | 深圳市欢太科技有限公司 | Video processing method and device, electronic equipment and computer readable storage medium |
| CN113723472A (en) * | 2021-08-09 | 2021-11-30 | 北京大学 | Image classification method based on dynamic filtering equal-variation convolution network model |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0601878A1 (en) * | 1992-12-10 | 1994-06-15 | Sony Corporation | Adaptive dynamic range encoding method and apparatus |
| US6323905B1 (en) * | 1997-12-25 | 2001-11-27 | Sony Corporation | Picture conversion apparatus picture conversion method learning apparatus and learning method |
| CN101325655A (en) * | 2007-06-15 | 2008-12-17 | 索尼株式会社 | Image signal processing apparatus and method, image display and output apparatus |
| US20090097542A1 (en) * | 2006-03-31 | 2009-04-16 | Sony Deutschland Gmbh | Signal coding and decoding with pre- and post-processing |
| US20090161947A1 (en) * | 2007-12-21 | 2009-06-25 | Sony Corporation | Image processing device and method, learning device and method, program, and recording medium |
| CN102413330A (en) * | 2007-06-12 | 2012-04-11 | 浙江大学 | Texture-adaptive video coding/decoding system |
| CN103475876A (en) * | 2013-08-27 | 2013-12-25 | 北京工业大学 | Learning-based low-bit-rate compression image super-resolution reconstruction method |
-
2015
- 2015-02-04 CN CN201510058184.1A patent/CN104661023B/en not_active Expired - Fee Related
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0601878A1 (en) * | 1992-12-10 | 1994-06-15 | Sony Corporation | Adaptive dynamic range encoding method and apparatus |
| US6323905B1 (en) * | 1997-12-25 | 2001-11-27 | Sony Corporation | Picture conversion apparatus picture conversion method learning apparatus and learning method |
| US20090097542A1 (en) * | 2006-03-31 | 2009-04-16 | Sony Deutschland Gmbh | Signal coding and decoding with pre- and post-processing |
| CN102413330A (en) * | 2007-06-12 | 2012-04-11 | 浙江大学 | Texture-adaptive video coding/decoding system |
| CN101325655A (en) * | 2007-06-15 | 2008-12-17 | 索尼株式会社 | Image signal processing apparatus and method, image display and output apparatus |
| US20090161947A1 (en) * | 2007-12-21 | 2009-06-25 | Sony Corporation | Image processing device and method, learning device and method, program, and recording medium |
| CN103475876A (en) * | 2013-08-27 | 2013-12-25 | 北京工业大学 | Learning-based low-bit-rate compression image super-resolution reconstruction method |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112805990A (en) * | 2018-11-15 | 2021-05-14 | 深圳市欢太科技有限公司 | Video processing method and device, electronic equipment and computer readable storage medium |
| CN110149554A (en) * | 2019-05-31 | 2019-08-20 | Oppo广东移动通信有限公司 | Method, apparatus, electronic equipment and the storage medium of video image processing |
| CN110149554B (en) * | 2019-05-31 | 2021-06-15 | Oppo广东移动通信有限公司 | Video image processing method and device, electronic equipment and storage medium |
| CN110598584A (en) * | 2019-08-26 | 2019-12-20 | 天津大学 | Convolutional neural network face recognition algorithm based on wavelet transform and DCT |
| CN113723472A (en) * | 2021-08-09 | 2021-11-30 | 北京大学 | Image classification method based on dynamic filtering equal-variation convolution network model |
| CN113723472B (en) * | 2021-08-09 | 2023-11-24 | 北京大学 | Image classification method based on dynamic filtering constant-variation convolutional network model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104661023B (en) | 2018-03-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107197260A (en) | Video coding post-filter method based on convolutional neural networks | |
| CN104661023B (en) | Image or method for video coding based on predistortion and training wave filter | |
| CN113055674B (en) | Compressed video quality enhancement method based on two-stage multi-frame cooperation | |
| US7697611B2 (en) | Method for processing motion information | |
| CN105898300A (en) | Improved transformation coefficient sign data hiding method based on recovered transformation coefficient | |
| CN101668196A (en) | Low code rate image compression method based on down sampling and interpolation | |
| CN113592746B (en) | Method for enhancing quality of compressed video by fusing space-time information from coarse to fine | |
| Wang et al. | Compressed screen content image super resolution | |
| CN115442613A (en) | Interframe information-based noise removal method using GAN | |
| Liu et al. | Convolutional neural network-based residue super-resolution for video coding | |
| CN108933942B (en) | Filtering method of compressed video and filtering device for compressed video | |
| CN105141967A (en) | Fast adaptive loop filter algorithm based on just noticeable distortion (JND) model | |
| Qi et al. | CNN-based post-processing filter for video compression with multi-scale feature representation | |
| JP6174966B2 (en) | Image coding apparatus, image coding method, and program | |
| Zhang et al. | WCDANN: A Lightweight CNN Post-Processing Filter for VVC-Based Video Compression | |
| Ding et al. | CARNet: Compression Artifact Reduction for Point Cloud Attribute | |
| Man et al. | Deep learning-assisted video compression framework | |
| Ning et al. | Video Reversible Data Hiding Based on Motion Vector | |
| CN113507607B (en) | Compressed video multi-frame quality enhancement method without motion compensation | |
| CN103841421A (en) | USDC data processing method | |
| Darazi et al. | Adaptive lifting scheme-based method for joint coding 3D-stereo images with luminance correction and optimized prediction | |
| Tang et al. | Chaotic moving video quality enhancement based on deep in-loop filtering | |
| CN114554213B (en) | Motion adaptive and detail-focused compressed video quality enhancement method | |
| Nayan et al. | Baseline JPEG-like DWT CODEC for disparity compensated residual coding of stereo images | |
| CN115496683A (en) | Compressed video quality enhancement method based on holoknowledge network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180309 Termination date: 20210204 |