CN108111852B - Double-measurement-parameter rate-distortion control method for quantization block compressed sensing coding - Google Patents
Double-measurement-parameter rate-distortion control method for quantization block compressed sensing coding Download PDFInfo
- Publication number
- CN108111852B CN108111852B CN201810031831.3A CN201810031831A CN108111852B CN 108111852 B CN108111852 B CN 108111852B CN 201810031831 A CN201810031831 A CN 201810031831A CN 108111852 B CN108111852 B CN 108111852B
- Authority
- CN
- China
- Prior art keywords
- measurement
- code rate
- measurement parameter
- rate
- 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.)
- Active
Links
Images
Classifications
-
- 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
- H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
- H04N19/146—Data rate or code amount at the encoder output
- H04N19/147—Data rate or code amount at the encoder output according to rate distortion criteria
-
- 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
- H04N19/10—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
- H04N19/132—Sampling, masking or truncation of coding units, e.g. adaptive resampling, frame skipping, frame interpolation or high-frequency transform coefficient masking
Abstract
The invention relates to a double-measurement parameter rate-distortion control method for quantization block compressed sensing coding, which takes a sampling rate and a quantization depth as measurement parameters, realizes the maximization of coding quality under the code rate constraint and is divided into a training mode and a working mode: under a training mode, restoring a target image under various measurement parameter values to construct a dual-measurement parameter model used by the controller; and under the working mode, distributing measurement parameter values according to the double-measurement parameter model, and performing quantization blocking compressed sensing coding on the target image so as to obtain optimized coding quality under the given code rate constraint. The invention can preset optimized measurement parameter values for subsequent target images, gradually adjust the coding quality under the constraint of different code rate levels, and can provide optimized rate distortion performance for the subsequent target images because adjacent target images have relatively consistent statistical characteristics.
Description
Technical Field
The invention relates to the technical field of image processing and coding, in particular to a double-measurement-parameter rate-distortion control method for quantization block-oriented compressed sensing coding.
Background
Compressed sensing enables the decomposition of a target signal into a small number of linear measurements, while enabling acquisition and compression of the signal, at a sampling rate well below the nyquist frequency. As a typical image compressed Sensing architecture, Quantized Block Compressed Sensing (QBCS) coding divides a target image into some non-overlapping blocks of the same size, and uses the same observation matrix to sequentially perform independent compressed sampling on each image Block, that is, perform random observation at an equal sampling rate on each Block, and then perform prediction and quantization coding on measured values, thereby generating a digitized code stream. At the QBCS measuring end, the scale of the observation matrix is not increased along with the increase of the size of the target image, so that the calculation and storage cost is reduced, and a low-power-consumption scheme is provided for transmitting the high-resolution image in real time. If the coding quality of the target image needs to be adjusted, the measuring end only needs to change the sampling rate of the observation matrix, so that the observation matrix is prevented from being changed in a large scale, and convenience is provided for hardware design.
The main goal of QBCS coding is to recover the target image with high quality by a lower code rate. The rate distortion performance of the current target image can be jointly controlled by the sampling rate and the quantization depth of the measuring end. Because the measuring end is difficult to obtain the content characteristics of different blocks through repeated measurement, the block self-adaptive parameter adjustment mechanism is difficult to be practically applied, and therefore, the fact that the same target image is subjected to the consistent sampling rate and the consistent quantization depth is more practical. However, under the constraint of a certain code rate, how to preset an optimized sampling rate and quantization depth for a target image to realize rate-distortion control of the QBCS coding quality and the code rate still lacks a related method at present.
Disclosure of Invention
The invention aims to solve the technical problem of providing a double-measurement-parameter rate-distortion control method for quantization block compressed sensing coding, which can provide optimized rate-distortion performance for subsequent target images.
The technical scheme adopted by the invention for solving the technical problems is as follows: the method for controlling the rate distortion of the double measurement parameters for the quantization block-oriented compressed sensing coding is provided, the sampling rate and the quantization depth are used as the measurement parameters, the maximization of the coding quality is realized under the code rate constraint, and the method comprises the following steps: under a training mode, restoring a target image under various measurement parameter values to construct a dual-measurement parameter model used by the controller; in the working mode, measurement parameter values are distributed according to the double-measurement parameter model, and QBCS coding is performed on the target image so as to obtain optimized coding quality under the given code rate constraint.
The method for constructing the double-measurement parameter model used by the controller comprises the following steps:
firstly, acquiring one or more original images of a target image, taking a sampling rate S and a quantization depth D as two independent measurement parameters, selecting all combinations of the sampling rate S and the quantization depth D, and respectively executing QBCS coding and QBCS decoding on the target image based on each group of measurement parameter values (S, D); after testing all the combined measurement parameter values (S, D) of the target image, all the coding quality Q (S, D) and the code rate R (S, D) of the target image are obtained.
For a given code rate level RtWhen the code rate R (S, D) is less than or equal to RtUnder the condition of (1), each coding quality Q (S, D) is searched in a traversing way, and a group of optimal measurement parameter values are respectively found in all possible measurement parameter values (S, D) for each target image, so that a code rate level R is obtainedtCorresponding optimal S distribution and optimal D distribution; after testing each code rate level R one by onetAfter the values are obtained, the model training unit calculates the optimal average value of S and the optimal average value of D for each code rate level, and curve fitting is carried out on the discrete data set; based on least square regression, selecting a general function form with the minimum mean square error and parameters thereof to form two fitting functions of the double-measurement parameter model: s (R)t) And D (R)t) The function values are limited to the maximum sampling rate and the maximum quantization depth, respectively.
The code rate level RtIt is varied in a certain step size within a possible range of values or from a given set of code rate levels.
The two fitting functions of the double-measurement parameter model are respectively as follows: s (R)t)＝ω1·Rt+ω2And
wherein, ω is1、ω2、τ1、τ2、τ3、τ4Are the model coefficients.
In the operating mode, according to a given code rate level RtAnd setting appropriate measurement parameter values for the subsequent target image by using the double-measurement-parameter model (S, D).
Advantageous effects
Due to the adoption of the technical scheme, compared with the prior art, the invention has the following advantages and positive effects: the invention takes the sampling rate and the quantization depth as the measurement parameters, and a double-measurement parameter model is fitted between the code rate level and the optimal parameters thereof through a training mode; based on the model, the invention adaptively allocates the measurement parameter values for the subsequent target image through the working mode so as to obtain the optimized coding quality under the code rate constraint. The invention provides an image-level rate-distortion control method for a target image to be coded by orienting to QBCS coding, can obtain optimized performance compromise between code rate and coding quality, and has certain expandability in the aspects of model application and code rate selection.
Drawings
Fig. 1 is a diagram of the functional elements of a dual measurement parameter rate-distortion control.
Fig. 2 is a diagram illustrating an encoding quality Q (S, D) curve of a target image.
FIG. 3 is a diagram illustrating the R (S, D) curve of the code rate of the target image.
Fig. 4 is a distribution diagram of the optimal sampling rate S at various code rates.
Fig. 5 is a distribution diagram of the optimal quantization depth D at various code rates.
FIG. 6 is S (R) of the dual measured parameter modelt) And (6) fitting a graph with the function.
FIG. 7 is D (R) of the dual measured parameter modelt) And (6) fitting a graph with the function.
Detailed Description
The invention will be further illustrated with reference to the following specific examples. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention may be made by those skilled in the art after reading the teaching of the present invention, and such equivalents may fall within the scope of the present invention as defined in the appended claims.
The embodiment of the invention relates to a dual-measurement-parameter rate-distortion control method for quantization block compressed sensing coding, wherein a QBCS (quantization block compression sensing coding) measurement end has two important measurement parameters: a sampling rate (S) and a quantized depth (D). If QBCS coding is performed on the same target image, a higher sampling rate S or quantization depth D will result in a higher code rate and higher coding quality. In the training mode of the embodiment, the model training unit performs all possible measurement parameter values (S, D) on a plurality of target images, and counts the influence of different parameters on QBCS coding quality and code rate, so as to construct a dual-measurement-parameter model. In the working mode of the embodiment, the controller allocates measurement parameter values (S, D) to the subsequent target image by using the dual-measurement parameter model according to the code rate level, and performs QBCS encoding. At the QBCS measurement side, a target image is divided into non-overlapping blocks of b x b pixels, all blocks having consistent measurement parameters, each block passing through an observation matrix phiBThe measured values are formed, and the number of measured values per block is b × b × S. FIG. 1 shows a functional unit diagram for implementing dual-measurement parameter rate-distortion control, wherein a QBCS measurement end comprises functional units such as model training, a controller, block division, random observation, quantization, entropy coding and the like; in the figure, xjRepresenting the jth block, X, in the target image XjBy the same observation matrix phiBIndependently observed to generate measured value yjY represents the measured value of all blocks of the whole image; since the QBCS measurements are real, they will be quantized to index values
After quantization, the measurement end performs entropy coding on the QBCS index value, and then encapsulates the compressed data and the auxiliary information into a code stream Y for digital storage or transmission.
Sampling rate S and quantization depth D jointly influence target mapThe coding quality and the code rate of the image. Q (S, D) represents the coding quality of a certain target image, and can be measured by a peak signal-to-noise ratio (PSNR), and the unit is dB; r (S, D) represents the bitrate of the target image in bits/pixel (bpp). For QBCS coding of a target image, a set of measured parameter values (S, D) can yield a pair of coding quality and code rate combinations: { Q (S, D), R (S, D) }. The goal of rate-distortion optimization is at the rate level RtMaximizing the coding quality Q (S, D) under constraints, forms the following two-parameter assignment problem:
maxQ(S,D)s.t.R(S,D)≤Rt(1)
in order to solve the above problem, the model training unit first tests the statistical properties of the target image under different measurement parameter values. The sampling rate S and the quantization depth D are two independent measurement parameters, and the measurement end selects different combinations of S (from 0.05 to the maximum sampling rate 1.00, step size 0.05) and D (from 0.5 to the maximum quantization depth 8.0, step size 0.5), which are 20 × 16 sets of measurement parameter values (S, D) in total in this embodiment. The model training unit first acquires an original image of a target image. In order to obtain QBCS (QBCS) coding characteristics of a target image, a measuring end respectively executes various measuring parameter values (S, D) on a plurality of target images, corresponding coding quality and code rate are counted, and a model training unit obtains a performance curve of the coding quality Q (S, D) and the code rate R (S, D) of the target images. Fig. 2 shows a graphical representation of the coding quality Q (S, D) in which a series of curves are obtained by varying the sampling rate S or the quantization depth D. When S is 1.0 and D is 8.0, Q (S, D) should be equal to its maximum value. Similarly, fig. 3 shows a graph of the code rate R (S, D). As can be seen from the figure, when the sampling rate S or the quantization depth D increases, the coding quality or the code rate increases, and the target image can achieve better coding quality at the cost of the increased code rate, which necessitates the selection of an optimization for the sampling rate and the quantization depth.
Aiming at a plurality of target images, the model training unit exhaustively searches the highest coding quality under each code rate level by inputting each pair of coding quality and code rate into the target function of the formula (1), and obtains the corresponding optimal S distribution or optimal D distribution. By testing several target images, fig. 4 shows the distribution of the optimal sampling rate S at various code rates, where one cross in the graph represents the combination of the optimal sampling rate S and its code rate level for a certain target image. Fig. 5 shows the distribution of the optimal quantization depths D at various code rates, and a plus sign in the figure indicates the combination of the optimal quantization depth of a certain target image and the code rate level thereof.
Based on least squares regression, the model training unit tests and evaluates typical function forms as much as possible, each code rate level RtThe optimal average of the corresponding sampling rate S or quantization depth D varies monotonically with the code rate level, respectively, so that curve fitting can be performed. Model training unit at code rate level R by fitting optimal mean of sampling rate S or quantized depth DtAnd a double-measurement parameter model is constructed between the measurement parameter values (S, D), and comprises two relation functions: s (R)t) Function sum D (R)t) A function. Optimal mean and code rate level R at StIn between, FIG. 6 shows S (R)t) And in the curve fitting process of the function, under a certain code rate, each cross mark represents the optimal average value based on S, and the dotted line represents the fitted function. Optimal mean and rate level R at DtFig. 7 shows D (R)t) And in the curve fitting process of the function, under a certain code rate, each plus sign represents the optimal average value based on D, and the dotted line represents the fitted function. By fitting the optimal mean value of S or D, the model training unit constructs a dual-measurement-parameter model comprising S (R) as followst) Function sum D (R)t) Function:
S＊(Rt)＝min{ω1·Rt+ω2，1.00} (2)
in the above formula, min { } represents that the minimum value is selected and should be smaller than the maximum value allowed by the parameter; rtRepresenting one code rate level and the other symbols representing the model coefficients. The values of the coefficients of the dual measured parameter model of the example are given in table 1, estimated by least squares regression.
TABLE 1 Dual measurement parameter model coefficient Table
The adjacent target images tend to have similar optimal sampling rates or optimal quantization depths at a given code rate, and the dual-measurement parameter model reveals an internal relation between the coding quality-code rate and the measurement parameter values. For a given code rate RtAnd the measuring end can provide optimized measuring parameter values (S, D) for the subsequent target image by using the equations (2) and (3), and the coding quality is gradually improved along with the increase of the code rate level. The invention can effectively distribute the parameters of the measuring terminal under certain code rate constraint, and obtain relatively optimized quality-code rate performance.
In the invention, the sampling rate and the quantization depth are used as measurement parameters, and a double-measurement parameter model is fitted between the code rate level and the optimal parameter of the code rate level through a training mode; based on the model, the invention adaptively allocates the measurement parameter values for the subsequent target image through the working mode so as to obtain the optimized coding quality under the code rate constraint. The invention provides an image-level rate-distortion control method for a target image to be coded by orienting to QBCS coding, can obtain optimized performance compromise between code rate and coding quality, and has certain expandability in the aspects of model application and code rate selection.
Claims (4)
1. A dual-measurement parameter rate-distortion control method for quantization block-oriented compressed sensing coding is characterized in that a sampling rate and a quantization depth are used as measurement parameters, the maximization of coding quality is realized under the code rate constraint, and the method is divided into a training mode and a working mode:
under a training mode, a target image is restored under various measurement parameter values, and a dual-measurement parameter model used by a controller is constructed, and the method specifically comprises the following steps:
firstly, acquiring original images of a plurality of target images, taking a sampling rate S and a quantization depth D as two independent measurement parameters, selecting all combinations of the sampling rate S and the quantization depth D, and respectively executing quantization blocking compressed sensing coding and decoding on the target images based on each set of measurement parameter values (S, D); after testing all the measurement parameter values (S, D) of the target image, obtaining all the coding quality Q (S, D) and the code rate R (S, D) of the target image;
for a given code rate level RtWhen the code rate R (S, D) is less than or equal to RtUnder the condition of (1), each coding quality Q (S, D) is searched in a traversing way, and a group of optimal measurement parameter values (S, D) are respectively found out from possible measurement parameter values (S, D) for each target image, so that a code rate level R is obtainedtCorresponding optimal S distribution and optimal D distribution; after testing each code rate level R one by onetAfter the values are obtained, the model training unit calculates the optimal average value of S and the optimal average value of D for each code rate level, and curve fitting is carried out on the discrete data set; based on least square regression, selecting a general function form with the minimum mean square error and parameters thereof to form two fitting functions of the double-measurement parameter model: s*(Rt) And D*(Rt) The function values are limited by the maximum sampling rate and the maximum quantization depth respectively;
and under the working mode, distributing measurement parameter values according to the double-measurement parameter model, and performing quantization blocking compressed sensing coding on the target image so as to obtain optimized coding quality under the given code rate constraint.
2. The dual-measurement-parameter rate-distortion control method for quantization-oriented block-wise compressed perceptual coding according to claim 1, wherein the rate level R istIt is varied in a certain step size within a possible range of values or from a given set of code rate levels.
3. The method according to claim 1, wherein the two fitting functions of the dual-measurement-parameter model are respectively: s＊(Rt)＝ω1·Rt+ω2And
wherein, ω is1、ω2、τ1、τ2、τ3、τ4Are the model coefficients.
4. The dual-measurement-parameter rate-distortion control method for quantization-oriented block-partitioning compressed sensing coding according to claim 1, wherein in the working mode, the rate-distortion control method is based on a given code rate level RtAnd setting appropriate measurement parameter values for the subsequent target image by using the double-measurement-parameter model (S, D).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810031831.3A CN108111852B (en) | 2018-01-12 | 2018-01-12 | Double-measurement-parameter rate-distortion control method for quantization block compressed sensing coding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810031831.3A CN108111852B (en) | 2018-01-12 | 2018-01-12 | Double-measurement-parameter rate-distortion control method for quantization block compressed sensing coding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108111852A CN108111852A (en) | 2018-06-01 |
| CN108111852B true CN108111852B (en) | 2020-05-29 |
Family
ID=62219968
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810031831.3A Active CN108111852B (en) | 2018-01-12 | 2018-01-12 | Double-measurement-parameter rate-distortion control method for quantization block compressed sensing coding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108111852B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110225340B (en) * | 2019-05-31 | 2021-07-09 | 北京猿力未来科技有限公司 | Control method and device for video coding, computing equipment and storage medium |
| CN110838151B (en) * | 2019-11-12 | 2020-07-03 | 南京甄视智能科技有限公司 | Picture compression processing method, computer system and readable storage medium |
| CN111669591B (en) * | 2020-05-09 | 2021-06-11 | 东华大学 | Frame group sampling rate distribution system for compressed sensing video stream |
| CN113422950B (en) * | 2021-05-31 | 2022-09-30 | 北京达佳互联信息技术有限公司 | Training method and training device, image data processing method and device, electronic device, and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2076048A2 (en) * | 2007-12-21 | 2009-07-01 | Samsung Electronics Co., Ltd. | Method, Medium, and Apparatus Representing Adaptive Information of 3D Depth Image |
| CN103124347A (en) * | 2012-10-22 | 2013-05-29 | 上海大学 | Method for guiding multi-view video coding quantization process by visual perception characteristics |
-
2018
- 2018-01-12 CN CN201810031831.3A patent/CN108111852B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2076048A2 (en) * | 2007-12-21 | 2009-07-01 | Samsung Electronics Co., Ltd. | Method, Medium, and Apparatus Representing Adaptive Information of 3D Depth Image |
| CN103124347A (en) * | 2012-10-22 | 2013-05-29 | 上海大学 | Method for guiding multi-view video coding quantization process by visual perception characteristics |
Non-Patent Citations (2)
| Title |
|---|
| 压缩感知的量化率失真分析;张旭坤 ，马社祥;《计算机应用》;20130101;全文 * |
| 率失真优化的压缩感知图像编码;蒋伟，杨俊杰;《电视技术》;20161130;说明书第13-16页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108111852A (en) | 2018-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108111852B (en) | Double-measurement-parameter rate-distortion control method for quantization block compressed sensing coding | |
| CN109495741B (en) | Image compression method based on self-adaptive down-sampling and deep learning | |
| KR100555601B1 (en) | Adaptive rate control for digital video compression | |
| CN102186085B (en) | Video processing method, video processing device and system | |
| CN103748876B (en) | Lossy compression method coded data method and apparatus and corresponding reconstruct data method and equipment | |
| CN104796705B (en) | A kind of HEVC Video codings rate-distortion optimization and rate control algorithm based on SSIM | |
| CN103051901B (en) | Video data coding device and method for coding video data | |
| Zhang et al. | Practical image quality metric applied to image coding | |
| CN104378636B (en) | A kind of video encoding method and device | |
| EP2654294A1 (en) | Wavelet coefficient quantization method applying human -eye visual model in image compression | |
| CN108574841A (en) | A kind of coding method and device based on adaptive quantizing parameter | |
| CN1463560A (en) | Method and system for determining distortion in video signal | |
| CN112203089B (en) | Image compression method, system and device based on code rate control of sparse coding | |
| CN1157079A (en) | Device and method for coding video pictures | |
| IL258485A (en) | Data compression | |
| US8351723B2 (en) | Methods and apparatus for generating and using visual error weights | |
| CN101472182B (en) | Virtually lossless video data compression | |
| CN101715143B (en) | Method, device and system of self-adaptive quantization and self-adaptive inverse quantization | |
| CN109361922B (en) | Predictive quantization coding method | |
| US7657110B2 (en) | Image compression using a color visual model | |
| CN109743571A (en) | A kind of image encoding method based on parallelly compressed perception multilayer residual error coefficient | |
| EP2243299B1 (en) | Method and device for compressing an image and storage medium carrying an image compressed by said method | |
| EP3180862B1 (en) | Method for coding pulse vectors using statistical properties | |
| CN111107363A (en) | Bandwidth compression coding method | |
| DE602004011213T2 (en) | INTRAFRAME COMPRESSION AND DECOMPRIMATION OF VIDEO SIGNALS WITH A FIXED BITRATE |
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 |