Image compression method based on JPEG-LS parallel optimization algorithm

Abstract

The invention discloses an image compression method based on a JPEG-LS (Joint Photographic Experts Group-Least Squares) parallel optimization algorithm, which comprises the following steps: calculating local gradient values of to-be-coded data and then merging to obtain a context index Q value, the to-be-coded data being a prediction error of a current pixel; grouping the data to be coded according to the context index Q value; if the to-be-coded number contained in the key group affects the parallelism degree, a controllable distortion value is introduced into the pixel value of the to-be-coded data by using an equalization algorithm, and the group serial number corresponding to the data in the key group is corrected; deploying processing units of parallel channels with the number equal to that of the groups, and scheduling the grouped data to be coded to the processing units to realize pipeline processing; integrating each group of data into coded and compressed data through code stream splicing; the image compression method based on the JPEG-LS parallel optimization algorithm is used for image compression, the system data size is low, and the satellite-ground transmission bandwidth pressure is small.

Description

Technical Field

[0001] This invention belongs to the field of synthetic aperture radar image processing and relates to an image compression method based on the JPEG-LS parallel optimization algorithm. Background Technology

[0002] With the iterative upgrades of platform technology, Synthetic Aperture Radar (SAR) systems have formed a multi-layered observation network encompassing spaceborne, airborne, and unmanned aerial vehicles. Compared to airborne or ground-based SAR systems, spaceborne SAR systems offer advantages in global coverage and long-term stable observation. However, the limited bandwidth of space-to-ground transmission is becoming increasingly prominent, resulting in direct transmission of raw data taking several hours, far exceeding the satellite overpass time, making compression a rigid requirement. Spaceborne SAR images, due to their wide swath and richly detailed ground features, demand high efficiency and high fidelity in compression technology. In the field of static image compression, the JPEG-LS algorithm, with its high fidelity and low-complexity coding architecture, is increasingly widely used in satellite remote sensing and other fields. Furthermore, its ease of hardware implementation makes it suitable for deployment on parallel computing platforms such as FPGA.

[0003] The JPEG-LS image compression algorithm is a low-complexity lossless or near-lossless image compression standard widely used for compressing static continuous-tone images. This algorithm has low computational complexity and is easy to implement in hardware. Currently, numerous JPEG-LS compressor designs based on FPGA and ASIC platforms have been proposed to achieve high energy efficiency and low cost. However, the context read-write conflict problem in the JPEG-LS algorithm leads to two dilemmas in the parallelization process of existing hardware solutions: first, it fails to fully exploit its parallelization potential; second, parallelization results in other defects such as decreased compression ratio and compromised compatibility. The register read-write conflict problem caused by dynamic context updates in the JPEG-LS image compression algorithm severely restricts the parallelization potential of image compression systems. Summary of the Invention

[0004] The purpose of this invention is to provide an image compression method based on the JPEG-LS parallel optimization algorithm, so as to solve the problem that the existing JPEG-LS image compression algorithm causes register read and write conflicts due to dynamic context updates, which seriously restricts the parallelization potential of the image compression system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An image compression method based on the JPEG-LS parallel optimization algorithm includes:

[0007] The local gradient values ​​of the data to be encoded are calculated and then merged to obtain the context index Q value. The data to be encoded is the prediction error of the current pixel.

[0008] The data to be encoded is grouped according to the context index Q value;

[0009] If the number of data to be encoded contained in the key group affects the parallelism p, a controllable distortion value is introduced into the pixel value of the data to be encoded using an equalization algorithm to correct the group number corresponding to the data in the key group. The key group is the group containing the most data to be encoded.

[0010] A processing unit with a parallel channel number equal to the number of groups is deployed. The data to be encoded after being corrected and grouped is scheduled to the processing unit to achieve pipelined processing. Each group of data is integrated into encoded and compressed data by splicing the code streams.

[0011] The beneficial effects of this invention are as follows: Addressing the limitation of parallelization in the JPEG-LS algorithm, a gradient dynamic grouping and controllable distortion equalization algorithm is proposed. The context access order is optimized through local pixel grouping and sorting, and controllable distortion is introduced to alleviate read / write conflicts, overcoming the single-pixel processing bottleneck of traditional encoders. The aim is to explore the optimal parallelization configuration under resource constraints, realizing a multi-channel parallel design for the JPEG-LS encoder. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of an image compression method based on the JPEG-LS parallel optimization algorithm according to an embodiment of the present invention.

[0014] Figure 2 This is a general framework diagram of the JPEG-LS encoder according to an embodiment of the present invention.

[0015] Figure 3 This is a flowchart of the traditional JPEG-LS normal encoding process.

[0016] Figure 4 This is a flowchart of the context modeling process according to an embodiment of the present invention.

[0017] Figure 5 This is a schematic diagram illustrating the reading of the following information according to an embodiment of the present invention.

[0018] Figure 6This is a schematic diagram of code stream grouping according to an embodiment of the present invention.

[0019] Figure 7 This is a waveform diagram of the JPEG-LS encoder function simulation according to an embodiment of the present invention.

[0020] Figure 8 This is the original image of the JPEG-LS compression effect comparison diagram of the embodiment of the present invention.

[0021] Figure 9 This is the original image of the JPEG-LS compression effect comparison diagram of the embodiment of the present invention.

[0022] Figure 10 This is a graph showing the SSIM statistical results under the high-speed parallel compression method of this invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] An image compression method based on the JPEG-LS parallel optimization algorithm according to an embodiment of the present invention, referencing... Figure 1 As shown, it mainly includes:

[0025] Step 1: Calculate the context index Q value of the data to be encoded.

[0026] In this invention, the data to be encoded refers to the prediction error of the current pixel. The context index Q value in this step is obtained by calculating and merging the local gradient values ​​of the data to be encoded. (Refer to...) Figure 4 The specific calculation process can be described as follows:

[0027] First, based on the location of the data to be encoded, the locations of its neighboring samples are determined. A causal module is then constructed to confirm the context. The method is as follows: the encoding of each sample value is performed by adjusting a small number of neighboring samples. Let x be the current location of the data to be encoded, and let a, b, c, and d be the four adjacent reconstructed samples, representing the adjacent encoded data to the left, above, upper left, and upper right of the data to be encoded, respectively. Their reconstructed values ​​are R. a R b R c R d ,like Figure 5 As shown.

[0028] Secondly, after completing the context verification, calculate the local gradient values ​​D1, D2, and D3 of the data to be encoded:

[0029]

[0030] When the local gradient values ​​D1, D2, and D3 are all less than or equal to the threshold NEAR, the data to be encoded is considered to have the same grayscale value as the neighboring cells, forming a flat region, and the system enters run-length encoding mode. If the local gradient values ​​D1, D2, and D3 are greater than the distortion threshold NEAR, the data to be encoded is considered to have different grayscale values ​​from the neighboring cells, forming a non-flat region, and the system enters normal encoding mode.

[0031] Then, the similar local gradient values ​​are quantized according to the threshold interval to obtain the quantized values ​​Q1, Q2, and Q3 of D1, D2, and D3.

[0032] For a grayscale image with a pixel depth of 8 bits, the quantization thresholds T1, T2, and T3 for the local gradient values ​​are set to 3, 7, and 21, respectively. Each Q... i It is divided into 9 intervals, and each interval is assigned an integer between -4 and 4, that is:

[0033]

[0034] Among them, Q i The local gradient value D i The corresponding gradient intensity level number for the region is D. i The corresponding quantization value.

[0035] Where T1, T2, and T3 are the quantization thresholds for local gradient values. For a grayscale image with an 8-bit pixel depth, the thresholds T1, T2, and T3 are 3, 7, and 21, respectively. Based on the relationship between D1, D2, D3, and the thresholds, the region number Q will be obtained after quantization. i Each Q i They are divided into 9 intervals, and each interval is assigned an integer between -4 and 4.

[0036] Finally, sign correction and merging of the local gradient quantization values ​​Q1, Q2, and Q3 are performed. (Q1, Q2, Q3) represents a context local gradient quantization vector when Q1, Q2, and Q3 satisfy any of the following conditions:

[0037] (1) Q1 < 0;

[0038] (2) Q1 = 0, Q2 < 0;

[0039] (3) Q1 = 0, Q2 = 0, Q3 < 0,

[0040] With the symbol SIGN set to -1, the context local gradient value quantization vector transformation is (-Q1, -Q2, -Q3).

[0041] If none of the above conditions are met, then the symbol SIGN is 1, and the context local gradient value quantization vector remains unchanged as (Q1, Q2, Q3).

[0042] Furthermore, after completing the sign correction, the local gradient value quantization vector needs to be converted into a single integer context index Q, with the conversion formula being: Q = |81Q1 + 9Q2 + Q3|.

[0043] Step 2: Group the data to be encoded according to the context index Q value.

[0044] Read-write conflicts typically occur between adjacent samples. Therefore, assuming the bitstream of the data to be encoded contains M adjacent samples, the M data to be encoded are divided into N groups according to the context index Q value, and the grouping formula is as follows:

[0045] g = G(Q) = 1 + (Q - 1) mod N

[0046] Where mod represents the modulo operation, and g is the group number corresponding to different data to be encoded;

[0047] Therefore, based on this grouping method, adjacent data to be encoded with the same Q value are grouped into the same group. The write-after-read problem is restricted to occurring within the group and will not occur between groups. Figure 6 As shown.

[0048] Step 3: If the number of data to be encoded contained in the key group affects the parallelism p, use the equalization algorithm to introduce a controllable distortion value into the pixel value of the data to be encoded and correct the group number corresponding to the data in the key group.

[0049] After grouping, the group containing the most data to be encoded is called the critical group. Since the critical group has the longest combinational logic latency, the system's parallel processing efficiency is determined by the number of data items (g) to be encoded within each critical group. m Therefore, p = M / g m This reflects the system's parallelism, where M is the total number of data to be encoded, and the parallelism p is affected by the number of data to be encoded within the key group, g. m Restrictions.

[0050] The correction method in this step mainly refers to, firstly, modifying the interruption condition of the run-length encoding mode, and changing the interruption condition I... x ≠R b Change to |I x -R b |>1, where I x R represents the pixel value corresponding to the location of the data to be encoded. b The reconstructed value of the data above the data to be encoded is obtained by introducing |I x -R bThe distortion value of 1 in |>1 allows more data to be encoded to enter the run-length encoding mode, increasing the amount of data entering the run-length encoding mode and reducing the amount of data entering the normal encoding mode, thereby reducing the number of times the write-after-read problem occurs.

[0051] Then, for the data to be encoded, x1 to x, which enter the normal encoding mode... M According to the context model, the data to be encoded is x1~x M The corresponding data in the previous row is b1~b M Calculate the data to be encoded x i+1 The corresponding context index Q i+1 and group number g i+1 :

[0052] Q i+1 =|81Q1(b i+2 -b i+1 )+9Q2(b i+1 -b i )+Q3(b i -x i )|

[0053] g i+1 =1+[(Q i+1 -1)modN]

[0054] Artificially introducing distortion values ​​to control x i For x′ i , so that Q(b) i -x i )≠Q(b i -x′ i ), where Q(b) i -x i ) is the original context index, Q(b) i -x′ i ) represents the context index after distortion has been corrected, x′ i pixel values The expression is as follows:

[0055]

[0056] in, These are the pixel values ​​of each data item to be encoded in this row. This is the reconstructed value from the previous row of data. The prediction error, i.e., the pixel value, is then calculated. Its predicted value P x The difference between the two values ​​is used to obtain the adjusted data to be encoded. The Q value is then recalculated based on the adjusted data, thus achieving Q(b). i -x i )≠Q(b i -x′ i).

[0057] Among them, the predicted value P x The calculation formula is as follows:

[0058]

[0059] Where R a R b R c These are the reconstructed values ​​of the adjacent encoded data to the left, above, and upper left of the data to be encoded, respectively.

[0060] After the above steps are performed, Q(b) is made i -x i )≠Q(b i -x′ i After that, use x′ i Replace x i Correspondingly, the corresponding group number g i+1 Change to g′ i+1 .

[0061] By using a load balancing algorithm, the number of data items to be encoded in the critical group can be reduced, thereby improving the parallelism of the pipeline processing. For all the data items to be encoded, x1 to x... M Perform traversal, using the counter cnt[g] i [For groups g1 to g] respectively N The data to be encoded within each group is counted to obtain the number of data items contained in each group. Simultaneously, the number of data items to be encoded within each key group is assigned to the register max_cnt. Then, all data items to be encoded from x1 to x... M Perform a traversal; if the number of key group data points is cnt[g] i ]>cnt[g′ i If ]+1, then use g′ i+1 Correct the group number g corresponding to the data in the key group. i+1 This ensures that the number of data items to be encoded within the key group is g. m Minimum.

[0062] The above process can be described by the following algorithm:

[0063]

[0064] Where x1~x M-1 This refers to the first to the (M-1)th original values ​​in the current data sequence to be encoded, used to calculate the correction value x′ of the current data to be encoded. i In equalization algorithms, the correction value for each piece of data to be encoded needs to be referenced to its previous original value; for example, the correction value for x... i x is needed at this time i-1Therefore, the values ​​of the first M-1 pixels need to be input to correct subsequent pixels. g2~g M This refers to the original group number of the 2nd to Mth data in the current data sequence to be encoded. It is used to detect inter-group conflicts. The algorithm determines whether the grouping needs to be adjusted by comparing the original group numbers of adjacent elements. Group number conflict detection needs to be compared with the group number of the previous data to be encoded, so it starts from the second data to be encoded.

[0065] Figure 2 The encoding framework of this invention is illustrated, wherein a preprocessing module is used to calculate the context index Q value and the corresponding group number g of the data to be encoded. An equalization module calculates and outputs the corrected group number g′ according to the given equalization algorithm rules. i and the corrected data to be encoded x′ i While RAM_A is being written to, the result from the equalization module is stored in RAM_A; when RAM_A is full, writing switches to RAM_B, while downstream modules read from RAM_A (ping-pong mechanism). For data to be encoded using normal encoding mode, the channel selection module allocates an encoding channel (normal encoding channel or run-length encoding channel) based on the group number corresponding to the data. When RAM_A is being written to, RAM_B is being read, and vice versa, achieving uninterrupted pipeline processing. The goal is to hide memory access latency and improve throughput (supporting processing multiple pixels per cycle), resulting in the corrected x′. i and group number g′ i Strict synchronization of output is required to ensure correct data allocation by the channel selection module. Ping-pong buffering and group number tagging enable conflict-free pipeline scheduling, supporting efficient parallel compression. The dual-RAM design avoids read / write conflicts in single-port RAM, while also saving resources. Finally, the stream integration module integrates the encoded data in the ping-pong RAM into an output of appropriate length.

[0066] Traditional methods such as Figure 3 As shown, in the JPEG-LS normal encoding mode flow, the local gradient value of the current pixel is first calculated to generate a context index, which is used for prediction and parameter updates. Then, the initial value A of the context parameter is used... o B o C o N o And the prediction error yields the final calculated result A of the context parameters. n B n C n N nThe calculation result stored in the register is assigned to the context parameter, and the assigned context parameter is used as the initial value for the next round of context parameter updates, thus completing one cycle of context parameter updates. In the prediction calculation, the optimal prediction value is selected based on the neighboring pixel values ​​through edge detection. In the error prediction stage, the difference between the actual pixel value and the predicted value is calculated. Finally, Golomb encoding is performed to encode the quantized error, using Golomb-Rice encoding to optimize the code length. JPEG-LS is based on pixel prediction and residual coding. By predicting the current pixel value and calculating the difference between it and the actual value (prediction error), the error is then encoded to reduce redundancy. Its core process includes context modeling, prediction, error correction, quantization, modulus reduction, and Golomb encoding. A key feature is dynamic context dependency—the current pixel encoding requires real-time reading and updating of the historical context state, leading to read-write conflicts when adjacent pixels share the same Q, becoming a bottleneck for parallelization.

[0067] Step 4: Deploy processing units with a parallel channel number equal to the number of groups, schedule the grouped data to be encoded to the processing units to achieve pipeline processing, and integrate each group of data into encoded and compressed data through bitstream splicing.

[0068] A numerical example illustrates the image compression effect of the image compression method based on the JPEG-LS parallel optimization algorithm according to this invention. The simulation parameters are: synthetic aperture radar transmitted signal pulse width 50μs, bandwidth 120MHz, carrier frequency 10.5GHz, pulse repetition frequency 7460Hz, signal-to-noise ratio 15dB, high-speed vehicle platform speed 7000m / s, and plasma sheath collision frequency 4GHz. When the peak electron density of the plasma sheath is 1×10⁻⁶... 17 / m 3 At that time, the partial signal waveform of the JPEG-LS encoder Figure 7 As shown in the figure, the i_x signal is the input data to be encoded, the o_bc signal is the output bitstream after Golomb encoding, and the o_bv signal is the frame assembled according to the JPEG-LS data protocol. The figure shows that the simulation waveform logic of the JPEG-LS encoder is correct. Figure 8 and Figure 9 As you can see, although the compressed image blurs the details, the ships and ports are still visually recognizable. Figure 10The histogram shown illustrates the Structure Similarity Index (SSIM) statistics for the images in the dataset used in this paper. Most images have an SSIM value greater than 0.998, with the worst SSIM value being 0.997 and an average SSIM of 0.99914. In the same benchmark test, at a compression ratio of 2.25, JPEG and JPEG2000 achieved average SSIM values ​​of 0.9965 and 0.9979, respectively. Compared to JPEG and JPEG2000, the efficient parallel JPEG-LS compression method proposed in this paper exhibits superior compression ratio-distortion balance characteristics in distortion-sensitive image compression applications, thus demonstrating the superiority of the image compression method based on the parallel optimization algorithm of JPEG-LS in this invention.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. An image compression method based on the JPEG-LS parallel optimization algorithm, characterized in that, include: The local gradient values ​​of the data to be encoded are calculated and then merged to obtain the context index Q value. The data to be encoded is the prediction error of the current pixel. The data to be encoded is grouped according to the context index Q value; If the number of data to be encoded contained in the key group affects the parallelism p, a controllable distortion value is introduced into the pixel value of the data to be encoded using an equalization algorithm to correct the group number corresponding to the data in the key group. The key group is the group containing the most data to be encoded. A processing unit with a parallel channel number equal to the number of groups is deployed. The data to be encoded after being corrected and grouped is scheduled to the processing unit to achieve pipelined processing. Each group of data is integrated into encoded and compressed data by splicing the code streams.

2. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 1, characterized in that, The context index Q value is calculated according to the following steps: First, after confirming the context, calculate the local gradient values ​​D1, D2, and D3 of the data to be encoded: When D1, D2, and D3 are all less than or equal to the threshold NEAR, the data to be encoded is considered to have the same gray value as the neighboring cells, forming a flat region, and the system enters run-length encoding mode; otherwise, the data to be encoded is considered to have different gray values ​​from the neighboring cells, forming a non-flat region, and the system enters normal encoding mode. Then, the similar local gradient values ​​are quantized according to the threshold interval to obtain the quantized values ​​Q1, Q2, and Q3 of D1, D2, and D3; Finally, sign correction is performed on Q1, Q2, and Q3. (Q1, Q2, Q3) represents a context local gradient quantization vector when Q1, Q2, and Q3 satisfy any of the following conditions: (1)Q1<0； (2) Q1 = 0, Q2 < 0; (3) Q1 = 0, Q2 = 0, Q3 < 0, With the symbol SIGN set to -1, the context local gradient value quantization vector transformation is (-Q1, -Q2, -Q3); If none of the above conditions are met, then the symbol SIGN is 1, and the context local gradient value quantization vector remains unchanged as (Q1,Q2,Q3); After completing sign correction, the context local gradient value quantization vector is converted into a single integer context index Q, with the conversion formula: Q = |81Q1 + 9Q2 + Q3|.

3. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 2, characterized in that, Based on the location of the data to be encoded, the locations of its neighboring samples are determined, a causal module is constructed, and the context confirmation is performed. The steps are as follows: Encoding of each sample value is performed by adjusting a small number of neighboring samples. Let x be the current position of the data to be encoded, and let a, b, c, and d be the four adjacent encoded samples to the left, above, upper left, and upper right of the data to be encoded, respectively, and their reconstructed values ​​are R. a R b R c R d .

4. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 2, characterized in that, For a grayscale image with a pixel depth of 8 bits, the quantization thresholds T1, T2, and T3 for the local gradient values ​​are set to 3, 7, and 21, respectively. Each Q... i It is divided into 9 intervals, and each interval is assigned an integer between -4 and 4, that is: Among them, Q i The local gradient value D i The corresponding gradient intensity level number for the region is D. i The corresponding quantization value.

5. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 1, characterized in that, The data to be encoded is grouped according to the context index Q value, following these steps: Suppose the bitstream of the data to be encoded contains M adjacent samples. Based on the context index Q value, the M data to be encoded are divided into N groups. The grouping formula is as follows: g = G(Q) = 1 + (Q - 1) mod N Where mod represents the modulo operation, and g is the group number corresponding to different data to be encoded; Accordingly, adjacent data to be encoded with the same Q value are grouped into the same group, thus limiting the write-after-read problem to occur within the group and preventing it from occurring between groups.

6. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 1, characterized in that, The formula for calculating the parallelism p is: p = M / g m , where g m M represents the number of data items to be encoded within the key group, where M is the total number of data items to be encoded.

7. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 1, characterized in that, The method of using an equalization algorithm to introduce controllable distortion values ​​into the pixel values ​​of the data to be encoded, and to correct the group number corresponding to the data within the key group, is as follows: First, the interruption condition of the run-length encoding mode is modified, and the interruption condition I is changed. x ≠R b Change to |I x -R b |>1, where I x R represents the pixel value corresponding to the location of the data to be encoded. b The reconstructed value of the data above the data to be encoded is obtained by introducing |I x -R b The distortion value of 1 in |>1 increases the amount of data entering run-length encoding mode and reduces the amount of data entering normal encoding mode; Then, for the data to be encoded, x1 to x, which enter the normal encoding mode... M According to the context model, the previous row of data corresponding to the data to be encoded is b1 to b2. M Calculate the data to be encoded x i+1 The corresponding context index Q i+1 and group number g i+1 Introducing distortion value control x i For x i ′, so that Q(b) i -x i )≠Q(b i -x i ′), where Q(b) i -x i ) is the original context index, Q(b) i -x i ′) represents the context index after correction by introducing distortion values; Using x′ i Replace x i The corresponding group number g i+1 Change to g′ i+1 ; For all the data to be encoded, x1 to x M Perform traversal, using the counter cnt[g] i [For groups g1 to g] respectively N The data to be encoded within the key group is counted to obtain the number of data in each group, and the number of data to be encoded in each key group is assigned to the register max_cnt. Then, all the data to be encoded, x1 to x2, were processed again. M Perform a traversal; if the number of key group data points is cnt[g] i ]>cnt[g′ i If ]+1, then use g′ i+1 Correct the group number g corresponding to the data in the key group. i+1 This ensures that the number of data items to be encoded within the key group is g. m Minimum.

8. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 7, characterized in that, Data to be encoded x i+1 The corresponding context index Q i+1 and group number g i+1 The calculation formula is as follows: Q i+1 =|81Q1(b i+2 -b i+1 )+9Q2(b i+1 -b i )+Q3(b i -x i )| g i+1 =1+[(Q i+1 -1)modN].

9. The image compression method based on the JPEG-LS parallel optimization algorithm according to claim 7, characterized in that, Through x i To x i The adjustment of ′ updates the local gradient value, making Q(b) more consistent with the expected value. i -x i )≠Q(b i -x i The specific steps are as follows: First, x i pixel value of ′ The expression is as follows: in, These are the pixel values ​​of each data item to be encoded in this row. This is the reconstructed value of the previous row of data; Then, the prediction error, i.e., the pixel value, is calculated. Its predicted value P x The difference between them is used to obtain the adjusted data to be encoded, where the predicted value P x The calculation formula is as follows: Where R a R b R c These are the reconstructed values ​​of the adjacent encoded data to the left, above, and above the left of the data to be encoded, respectively; Finally, the Q value is recalculated based on the adjusted data to be encoded, thus achieving Q(b) i -x i )≠Q(b i -x i ′).

Images