Multi-compatible high-fluency lossless decoding method and system

Abstract

The invention discloses a multi-compatible high-fluency lossless decoding method and system, and relates to the field of video processing, and the method comprises the steps: generating a candidate decoding sequence; obtaining required resource state parameters of each image block; obtaining the number of current image blocks capable of being decoded in parallel; and when the number of the parallel decoding image blocks is smaller than the threshold value of the number of the target parallel decoding image blocks, starting a speed-up compensation mechanism to perform virtual decoding. According to the method, a decoding sequence is generated through image block level division and topological sorting of a video image frame, and a parallel decoding decision-making mechanism under resource constraint is formed in combination with compression complexity and resource consumption parameters of image blocks. The problems of decoding delay, frame loss, processing congestion and the like are effectively avoided, and the real-time smooth decoding capability of the high-resolution video under multiple platforms is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of video processing, and in particular to a multi-compatible high-smooth lossless decoding method and system. Background Art

[0002] High-smooth lossless decoding achieves high frame rates, low latency, and smooth video decoding without sacrificing image or video quality. It combines the dual requirements of decoding quality and performance, making it suitable for scenarios requiring the highest quality and timeliness, such as 8K video playback, professional film and television editing, medical image analysis, and virtual reality.

[0003] Prior art, such as the invention patent with announcement number: CN114175641B, is about improving the efficiency of lossless coding and decoding in video coding and decoding. An electronic device performs a method for decoding video data in the following manner: receiving a first indication associated with a first segmentation level of the hierarchical structure from a video bitstream having a hierarchical structure; determining that the first indication indicates that a lossless mode is enabled at the first segmentation level: configuring one or more coding tools according to the lossless mode; and using the configured one or more coding tools to decode a codec block at or below the first segmentation level.

[0004] The prior art, such as the invention patent with announcement number CN114556932B, is a lossless codec mode for video coding and decoding, in which an electronic device performs a method for encoding and decoding video data. The method includes: receiving transform coefficients of a current codec block; scanning the transform coefficients to identify the last non-zero transform coefficient and its corresponding position in the current codec block, the position including an x-dimension and a y-dimension; selecting a first context model for the x-dimension from a context model group according to a first size of the current codec block along the x-dimension; selecting a second context model for the y-dimension from the context model group according to a second size of the current codec block along the y-dimension; encoding the x-dimension corresponding to the last non-zero transform coefficient into a video bitstream using the first context model; and encoding the y-dimension corresponding to the last non-zero transform coefficient into the video bitstream using the second context model.

[0005] Based on the above solutions, we can see that most existing image / video decoding methods rely on specific codec standards (such as H.264, HEVC, ProRes, and Notch LC). However, hardware acceleration support varies widely, and decoding logic is tightly coupled. In practical applications, whole-frame decoding is difficult to parallelize, resulting in low resource utilization. For videos requiring high-resolution decoding, I / O bandwidth and decoded frame reconstruction are burdened, leading to frequent lags. Summary of the Invention

[0006] In view of the shortcomings of the prior art, the present invention provides a multi-compatible, high-smoothness, lossless decoding method and system. To achieve the above objectives, the present invention is implemented through the following technical solutions: A multi-compatible, high-smoothness, lossless decoding method, comprising: Extract the video with high smoothness and lossless decoding requirement, divide the original image frame in the video with high smoothness and lossless decoding requirement into image blocks, perform topological scheduling on each image block of the original image frame, and generate a candidate decoding sequence.

[0007] Obtain the decoding cost characteristic value of each image block in the candidate decoding sequence, and map it to obtain the required resource status parameters of each image block. The required resource status parameters include the required CPU load ratio, the required GPU memory load ratio, the required system memory capacity ratio, and the required disk IO resource ratio.

[0008] Real-time hardware resource status parameters are obtained, and based on the candidate decoding sequence and the required resource status parameters of each image block, the number of image blocks that can currently be decoded in parallel is obtained.

[0009] A judgment is made based on the current number of image blocks that can be decoded in parallel. When the number of image blocks that can be decoded in parallel is less than the target threshold number of image blocks that can be decoded in parallel, a speed-up compensation mechanism is started to perform virtual decoding.

[0010] The original image frames in the video requiring high-smoothness lossless decoding are divided into image blocks, and topological scheduling is performed on each image block of the original image frame. The specific process is as follows: The required frame rate of the highly smooth lossless decoded video is obtained, the highly smooth lossless decoded video is separated into original image frames according to the required frame rate, and the original image frames are divided into image blocks based on a preset fixed granularity.

[0011] Image blocks are the basic scheduling units for decoding tasks.

[0012] The in-degree value of each image block is calculated, and candidate decoding sequences are generated from small to large based on the size of the in-degree value of each image block.

[0013] As a preferred technical solution, the candidate decoding sequence is generated in the following process: When generating candidate decoding sequences, if there are multiple image blocks with the same in-degree value in a certain original image frame, such image blocks are extracted and grouped into the same in-degree image block group. The position coordinates of each image block in the same in-degree image block group are obtained, and the Euclidean distance between each image block and the center point of the original image frame is calculated, which is recorded as the non-centrality.

[0014] The temporal change intensity indexes between each image block in the same in-degree image block group in the original image frame and the image block corresponding to the same position coordinates in the previous image frame are obtained, including PSNR gradient change, SSIM gradient change, optical flow intensity, MSE, edge map overlap and grayscale change. The temporal change intensity indexes of each image block in the same in-degree image block group are processed across image blocks to obtain the mean set of temporal change intensity indexes. As a reference value, the mean set is compared with the temporal change intensity indexes of each image block in the same in-degree image block group and weighted coupling processing is performed to obtain the change intensity value of each image block in the same in-degree image block group.

[0015] The non-centrality and variation intensity of each image block in the same in-degree image block group are comprehensively fitted to obtain the candidate decoding sequence value of each image block in the same in-degree image block group, and the candidate decoding sequence of each image block in the same in-degree image block group is obtained by sorting them from large to small.

[0016] As a preferred technical solution, obtaining the decoding cost characteristic value of each image block in the candidate decoding sequence specifically includes: The compression features of each image block in the candidate decoding sequence are analyzed and obtained, including the number of block compression bytes, entropy coding type and the number of non-zero coefficients of the residual block.

[0017] The entropy coding type of each image block is input into the mapping of entropy coding type-decoding cost reference amount pre-stored in the database, and the decoding cost reference amount corresponding to the entropy coding type of each image block is obtained by mapping and matching.

[0018] A set of compression feature reference values ​​is extracted from the database, including a reference value of the number of compressed bytes of a block and a reference value of the number of non-zero coefficients of a residual block.

[0019] The compression features of each image block are compared and analyzed with the compression feature reference value, and a weighted coupling process is performed. The decoding cost reference is introduced for comprehensive analysis to obtain the decoding cost characteristic value of each image block. The decoding cost characteristic value is used to quantify the size of the resources required for decoding the image block.

[0020] As a preferred technical solution, mapping obtains the required resource status parameters of each image block, specifically including: The decoding cost characteristic value of each image block is input into the mapping set of decoding cost characteristic value-required resource status parameter set pre-stored in the database, and mapping matching is performed to obtain the required resource status parameter set of each image block, including the required CPU load ratio, the required GPU video memory load ratio, the required system memory capacity ratio and the required disk IO resource ratio.

[0021] As a preferred technical solution, real-time hardware resource status parameters are obtained, and based on the candidate decoding sequence and the required resource status parameters of each image block, the number of image blocks that can currently be decoded in parallel is obtained, specifically including: Real-time hardware resource status parameters include the current remaining CPU load ratio, the current remaining GPU memory load ratio, the remaining system memory capacity ratio, and the current remaining disk IO resource ratio.

[0022] In the candidate decoding sequence, they are loaded into the decoding slot one by one in sequence, and their required resource status parameters are accumulated in real time until any real-time hardware resource status parameter is equal to its corresponding accumulated required resource status parameter. The loading is terminated, and the number of image blocks loaded in the decoding slot is the number of image blocks that can currently be decoded in parallel.

[0023] As a preferred technical solution, when the number of parallel decodable image blocks is less than the target parallel decodable image block number threshold, a speed-up compensation mechanism is activated to perform virtual decoding, specifically including: Based on the required frame rate of the required high-smoothness lossless decoded video, the decoding limit time of each original image frame is obtained, and the total number of image blocks of each original image frame is extracted to obtain the number of image blocks that need to be decoded within the decoding limit time.

[0024] The minimum resource status parameter among the real-time hardware resource status parameters is obtained, and the mapping of the corresponding resource status parameter-scheduling cycle number is input and mapped and matched to obtain the current scheduling cycle number.

[0025] The decoding limit time is divided into several scheduling cycles according to the number of scheduling cycles, and the number of image blocks that need to be decoded within the decoding limit time is evenly divided into each scheduling cycle to obtain the target parallel decoding image block number threshold of each scheduling cycle.

[0026] The number of parallel decodable image blocks in the decoding slot is monitored in real time. When the number of parallel decodable image blocks is less than the target parallel decodable image block number threshold, it is determined to be a decoding delay and the speed-up compensation mechanism is activated.

[0027] As the preferred technical solution, the speed-up compensation mechanism has the following specific processing conditions: The average change intensity value of the image blocks of each original image frame is extracted and compared with the average change intensity threshold preset in the database. If the average change intensity of the image blocks of an original image frame is less than the average change intensity threshold, the original image frame is determined to be a low change intensity image frame.

[0028] If the average change intensity of the image blocks of an original image frame is greater than or equal to the average change intensity threshold and less than the change intensity significance value, the original image frame is determined to be a medium change intensity image frame.

[0029] If the average change intensity of the image blocks of an original image frame is greater than or equal to the change intensity significance value, the original image frame is determined to be a high change intensity image frame. For high change intensity image frames, the normal decoding state is still maintained.

[0030] The average change intensity difference between the average change intensity value of the low change intensity image frame and the average change intensity threshold is extracted, and the mapping of the average change intensity difference-repair derivation factor pre-stored in the database is input for mapping matching to obtain the repair derivation factor of the low change intensity image frame, and the low change intensity image frame is virtually decoded.

[0031] The significant difference in change intensity between the average change intensity value and the significant change intensity value of the medium change intensity image frame is extracted, and the mapping of the significant change intensity difference value-virtual decoded image block ratio pre-stored in the database is input for centralized mapping matching to obtain the virtual decoded image block ratio of the medium change intensity image frame, and partial virtual decoding is performed on the medium change intensity image frame.

[0032] As a preferred technical solution, virtual decoding specifically includes: The image blocks of the previous frame of the low-variation intensity image frame are extracted and corresponded one-to-one with the image blocks of the low-variation intensity image frame to perform image block placeholders. The restoration derivation factors are input into the restoration derivation factor-complete virtual frame parameter mapping set preset in the database for mapping and matching to obtain the complete virtual frame parameters of the low-variation intensity image frame, including pixel interpolation and illumination derivation values. Image restoration is performed on the low-variation intensity image frame to complete the virtual decoding of the low-variation intensity image frame.

[0033] An image block of a medium-varying intensity image frame is extracted. Based on the candidate decoding sequence and the ratio of virtual decoded image blocks of the medium-varying intensity image frame, the position coordinates of the virtual decoded image block in the medium-varying intensity image frame are obtained. The previous image frame of the medium-varying intensity image frame is extracted. The image block with the same position coordinates as the virtual decoded image block is extracted. After one-to-one correspondence, the image block is occupied, thereby completing partial virtual decoding of the medium-varying intensity image frame.

[0034] In addition, a multi-compatible high-smooth lossless decoding system is also provided, including: The candidate decoding sequence module is used to extract the video with high smoothness and lossless decoding requirements, divide the original image frames in the video with high smoothness and lossless decoding requirements into image blocks, perform topological scheduling on each image block of the original image frame, and generate a candidate decoding sequence.

[0035] The decoding cost calculation module is used to obtain the decoding cost characteristic value of each image block in the candidate decoding sequence and map it to obtain the required resource status parameters of each image block. The required resource status parameters include the required CPU load ratio, the required GPU memory load ratio, the required system memory capacity ratio and the required disk IO resource ratio.

[0036] The parallel decoding module is used to obtain real-time hardware resource status parameters and obtain the number of image blocks that can be decoded in parallel based on the candidate decoding sequence and the required resource status parameters of each image block.

[0037] The virtual decoding module is used to make a judgment based on the current number of image blocks that can be decoded in parallel, and when the number of image blocks that can be decoded in parallel is less than the target number threshold of image blocks that can be decoded in parallel, start the speed-up compensation mechanism to perform virtual decoding.

[0038] Compared with the prior art, the embodiments of the present invention have at least the following beneficial effects: This invention provides a multi-compatible, highly smooth lossless decoding method. This method generates decoding sequences by partitioning and topologically sorting video frames at the image block level. This method, combined with image block compression complexity and resource consumption parameters, forms a resource-constrained, parallel decoding decision-making mechanism. This method not only supports adaptation to resource differences across multiple platforms but also effectively avoids issues such as decoding delays, frame drops, and processing congestion. It enables high-frame-rate, non-stuttering video playback and processing in high-load scenarios such as ultra-high-definition lossless video.

[0039] The present invention divides the original image frame into image blocks of fixed granularity, constructs a topological structure based on the image block dependencies, and generates candidate decoding sequences, thus achieving the evolution from image frame-level scheduling to image block-level scheduling. The decoding order is optimized based on in-degree calculation and non-centrality analysis, and the decoding throughput per unit time is significantly improved by decoding image blocks in parallel. At the same time, multi-dimensional sorting is performed based on the change intensity, decoding cost, and spatial location characteristics of the image blocks, which improves the spatial continuity and computational efficiency of the decoding scheduling, reduces the probability of decoding delay, and enhances the real-time and smooth decoding capabilities of high-resolution video on multiple platforms.

[0040] This paper introduces a real-time hardware resource status awareness mechanism, including CPU load, GPU memory usage, memory usage, and disk I / O resource usage. By comparing decoding cost feature values ​​with a pre-set mapping model in a database, it dynamically estimates the maximum number of parallel decoded image blocks the system can support. This mechanism ensures that decoding scheduling always operates within the safe range of the current device's resource capabilities, effectively avoiding system freezes, crashes, or low frame rates caused by resource overload.

[0041] In the face of insufficient resources, sudden increase in decoding pressure, or inability to complete decoding of specific frames in a timely manner, the present invention enables a speed-up compensation mechanism to guide some image blocks into the virtual decoding process according to the intensity of image block changes and scheduling cycle restrictions. By introducing strategies such as interpolation, illumination derivation, and placeholder multiplexing for low / medium intensity change image frames, the recalculation operation is partially or completely skipped, which not only ensures decoding efficiency but also does not significantly affect the image quality perceptible to the human eye. This hierarchical decoding scheme based on image content semantics greatly improves the flexibility and robustness of decoding tasks, ensures frame rate continuity and visual consistency, and is an important support mechanism for lossless video high frame rate playback scenarios.

[0042] Of course, any product implementing the present invention does not necessarily need to achieve all of the above advantages at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of the method of the present invention.

[0044] Figure 2 Schematic diagram of the system module of the present invention.

[0045] Figure 3 Schematic diagram of the logic flow involved in the embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] In the description of the present invention, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inside", "around" and the like indicating orientation or positional relationship are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0048] See also Figure 1 As shown, an embodiment of the present invention provides a multi-compatible high-smoothness lossless decoding method, including: like Figure 3 It is a schematic diagram of the logic flow involved in the embodiment of the present invention.

[0049] Extract the video with high smoothness and lossless decoding requirement, divide the original image frame in the video with high smoothness and lossless decoding requirement into image blocks, perform topological scheduling on each image block of the original image frame, and generate a candidate decoding sequence.

[0050] Extract the original frame rate information from the video encoding header to obtain the required frame rate for the smooth and lossless decoded video. Separate the smooth and lossless decoded video into original image frames according to the required frame rate, and divide the original image frames into image blocks based on a preset fixed granularity. The specific process includes: Each original image frame is divided into several equal-sized rectangular areas according to a preset fixed granularity. For example, an original image frame with an 8K resolution of 7680×4320 has a fixed granularity of 64×64 pixels. Each image block obtained after segmentation contains position coordinates and pixel area range. The image block is the basic scheduling unit of the decoding task.

[0051] The in-degree value of each image block is calculated, and candidate decoding sequences are generated from small to large based on the size of the in-degree value of each image block.

[0052] In-degree refers to how many other image blocks an image block depends on, that is, the number of upstream image blocks that need to be decoded before it can be decoded.

[0053] When generating candidate decoding sequences, if there are multiple image blocks with the same in-degree value in a certain original image frame, such image blocks are extracted and grouped into the same in-degree image block group. Since there is no mandatory order in the dependency relationship between these image blocks, their decoding order needs to be further screened.

[0054] The position coordinates of each image block in the same in-degree image block group are obtained, and the Euclidean distance between each image block and the center point of the original image frame is calculated, which is recorded as the non-centrality.

[0055] Obtain the temporal change intensity indicators between each image block in the same in-degree image block group in the original image frame and the image block corresponding to the same position coordinate in the previous image frame, including PSNR gradient change, SSIM gradient change, optical flow intensity, MSE, edge map overlap and grayscale change.

[0056] It should be noted that the PSNR (Peak Signal-to-Noise Ratio) gradient change is used to measure the similarity of two images in terms of overall brightness difference. The higher the value, the more similar they are and the smaller the change.

[0057] The SSIM (Structural Similarity Index) gradient change is a similarity index that is more consistent with human perception and takes into account multiple factors such as structure, brightness, and contrast. The range is generally [0, 1], where 1 indicates complete identity.

[0058] Optical flow intensity represents the motion vector (direction and speed) of each pixel in the video. Optical flow intensity is the modulus (speed) of the motion vector; larger values ​​indicate more intense motion within the image block. This is obtained using classic optical flow estimation algorithms such as Lucas-Kanade, Farneback, or DeepFlow.

[0059] MSE (Mean Squared Error) is used to measure the point-by-point difference between the pixels of two image blocks (regardless of structure). It is obtained by comparing the squared differences pixel by pixel in the same position and then averaging them.

[0060] The Edge Overlap Ratio evaluates the similarity in edge contours between the image blocks at the same location in the current and previous frames. A higher value indicates stable edges, while a lower value indicates significant changes. This is achieved by performing edge detection (e.g., Canny or Sobel) on the tile images of the current and previous frames, generating a binary edge map, and calculating the ratio of the intersection to the union of edge pixels.

[0061] Grayscale Delta refers to the change in average brightness within an image block between two frames. It is used to detect background changes such as large-area illumination, exposure, and shadows. To obtain this value, convert the tile image to grayscale, calculate the average grayscale value of the image block in the current and previous frames, and then take the absolute difference to obtain the grayscale delta.

[0062] The temporal change intensity index of each image block in the same in-degree image block group is processed across image blocks to obtain a set of temporal change intensity index means, which are used as reference values, including the PSNR gradient change mean, SSIM gradient change mean, optical flow intensity mean, MSE mean, edge map coincidence mean and grayscale change mean. They are compared with the temporal change intensity index of each image block in the same in-degree image block group and weighted coupled to obtain the change intensity value of each image block in the same in-degree image block group.

[0063] The non-centrality and variation intensity of each image block in the same in-degree image block group are subjected to comprehensive fitting processing to obtain candidate decoding sequence values ​​of each image block in the same in-degree image block group, and the candidate decoding sequences of each image block in the same in-degree image block group are obtained by sorting them from large to small, specifically including: ; ; Among them, LI i is the candidate decoding sequence value of the i-th image block in the same in-degree image block group, U i is the non-centrality of the i-th image block in the same in-degree image block group, Cha i is the change intensity value of the i-th image block in the same in-degree image block group, PSNR i is the PSNR gradient change of the i-th image block in the same in-degree image block group, SSIM i is the SSIM gradient change of the i-th image block in the same in-degree image block group, OFM iis the optical flow intensity of the i-th image block in the same in-degree image block group, MES i is the MSE and EOR of the i-th image block in the same in-degree image block group i GD is the edge map overlap of the i-th image block in the same in-degree image block group. i is the grayscale change of the i-th image block in the same in-degree image block group, is the mean value of PSNR gradient change, is the mean value of SSIM gradient change, is the mean optical flow intensity, is the mean MSE, is the mean overlap of edge graphs, is the mean grayscale change, α1 is the PSNR gradient change weighting factor, α2 is the SSIM gradient change weighting factor, α3 is the optical flow intensity weighting factor, α4 is the MSE weighting factor, α5 is the edge map coincidence weighting factor, α6 is the grayscale change weighting factor, i is the image block number in the same in-degree image block group, i=1,2,3,...,n, n is the number of image blocks in the same in-degree image block group.

[0064] It should be noted that the six temporal variation intensity indicators—PSNR gradient change, SSIM gradient change, optical flow intensity, MSE, edge map overlap, and grayscale change—each reflect the differential characteristics of image blocks in the time series from different dimensions. The PSNR (Peak Signal-to-Noise Ratio) gradient change measures the overall difference in brightness between the original image block and the image block at the same position in the previous frame. Higher values ​​indicate more similar images and smaller changes. This is particularly suitable for quickly identifying areas in image blocks with stable overall brightness and little content change. These areas are prioritized as candidates for complete virtual decoding. However, because PSNR does not focus on image structure, it may not accurately reflect the degree of change in scenes with slight contour movement or edge perturbations.

[0065] The Structural Similarity Index (SSIM) gradient variation combines image structure, brightness, and contrast to simulate the human eye's ability to perceive similarity in image content. This makes it a more realistic variation assessment metric. SSIM is highly sensitive to image edges, texture, and structural changes, making it suitable for use with PSNR to distinguish structural changes from non-structural brightness fluctuations.

[0066] Optical flow intensity, a metric describing pixel-level motion vectors, reflects the intensity of spatial motion within an image block, with larger values ​​indicating more dramatic changes. This metric is particularly sensitive to dynamic scenes such as moving objects and camera movements, making it a key parameter for determining whether high-fidelity decoding is necessary. Compared to grayscale changes or structural differences in static areas, optical flow intensity can identify visible motion.

[0067] MSE (Mean Squared Error) focuses on pixel-level grayscale differences, reflecting the point-by-point numerical changes within an image block. While it doesn't consider structure, it has a strong ability to detect low-level changes such as compression errors and small perturbations, making it a useful complement to SSIM.

[0068] Edge map overlap accurately characterizes object shapes and boundary changes by comparing the edge contour overlap of the current image block with the same block in the previous frame. This approach offers significant advantages in image segmentation, foreground object detection, and edge-preserving decoding, and is particularly effective for detecting areas with significant structural changes, such as occlusion and deformation.

[0069] Grayscale variation focuses on the average change in overall brightness across an image block and is primarily used to detect background changes such as large-area illumination, exposure, and shadows. This metric provides additional redundant information for addressing issues such as ambient light fluctuations, nighttime video, and lighting transitions.

[0070] It's important to note that the PSNR gradient change weighting factor, SSIM gradient change weighting factor, optical flow intensity weighting factor, MSE weighting factor, edge map overlap weighting factor, and grayscale change weighting factor all play a crucial role in controlling the importance of each metric in the fusion process, reflecting their perceived weighting over overall visual change. For example, the optical flow intensity weighting factor can be used to emphasize the importance of regions with dynamic motion, while the edge map overlap weighting factor helps highlight the significance of structural changes. In an embodiment of the present invention, the method for obtaining the PSNR gradient change weighting factor, the SSIM gradient change weighting factor, the optical flow intensity weighting factor, the MSE weighting factor, the edge map coincidence weighting factor and the grayscale change weighting factor is specifically a data-driven learning method, which specifically includes constructing a training set, introducing a deep learning model, and automatically learning the actual impact of each change indicator on the degree of perceptual change from a large amount of simulated video image data, thereby determining the weighting factor. The specific training process includes taking each temporal change intensity indicator (such as PSNR, SSIM, optical flow, etc.) in the simulated video image data as input features, training the deep learning model, and finally obtaining the weighting factor of each indicator based on the feature weight after model training.

[0071] It's also worth noting that there are certain correlations between PSNR gradient change, SSIM gradient change, optical flow intensity, MSE, edge map overlap, and grayscale variation. These parameters exhibit close interrelationships and significant complementary effects. First, there's a clear mathematical relationship between PSNR and MSE: PSNR is based on the inverse logarithm of MSE. As the pixel error (MSE) between images increases, the PSNR value decreases. The two represent different scales of the same change trend, thus exhibiting mathematical transformation dependence. Grayscale variation, as a fundamental pixel-level indicator of change, directly reflects the fluctuation in image brightness distribution and is the fundamental signal source for PSNR and SSIM. Therefore, it forms a transmission chain from underlying pixel changes to perceptual quality changes. Optical flow intensity describes the magnitude of motion of objects or pixels in an image sequence and is the primary source of dynamic information, while PSNR and SSIM emphasize the magnitude of changes in texture and structure. When optical flow intensity is significant while PSNR / SSIM changes slightly, it often indicates consistent image motion (such as background translation). Conversely, if all three are significant, it indicates significant changes in the regional content structure. Therefore, optical flow intensity and perceptual quality indicators form a dynamic-structural coupling relationship. Edge map coincidence focuses on the spatial matching of image structural contours. Like SSIM, it also focuses on image structure, resulting in a high correlation between the two, particularly in corroborating changes in edge-significant regions. In summary, these six metrics are interrelated and complementary across multiple dimensions, including numerical logic, perceptual hierarchy, dynamic characteristics, and structural preservation. Weighted fusion of these parameters effectively supports image block prioritization and resource allocation strategies during decoding scheduling.

[0072] Obtain the decoding cost characteristic value of each image block in the candidate decoding sequence, and map it to obtain the required resource status parameters of each image block. The required resource status parameters include the required CPU load ratio, the required GPU memory load ratio, the required system memory capacity ratio, and the required disk IO resource ratio.

[0073] Obtaining the decoding cost characteristic value of each image block in the candidate decoding sequence, specifically including: The compression features of each image block in the candidate decoding sequence are analyzed and obtained, including the number of block compression bytes, entropy coding type and the number of non-zero coefficients of the residual block.

[0074] The number of compressed bytes in a block is used to assess the data volume occupied by the image block in the bitstream. A larger number of bytes indicates richer information, but requires more decoding time and cache resources. This metric is usually determined by parsing the underlying bitstream. The entropy coding type indicates the entropy coding method used during the compression process for the image block. For example, CABAC (Context-Adaptive Binary Arithmetic Coding) is more complex than CAVLC (Context-Adaptive Variable Length Coding), resulting in greater decoding time and computational resource requirements. This metric is obtained by low-overhead reading of specific bit fields. The number of non-zero coefficients in the residual block reflects the complexity of the image block in the transform domain. This metric is directly related to the intensity of inverse quantization and inverse transform computations during decoding. A larger number of non-zero coefficients indicates more dramatic changes in the content of the image block, resulting in a heavier decoding computational burden. This metric is obtained by reverse parsing the transform coefficient syntax units (such as coeff_token and level_prefix) in the decoder.

[0075] The entropy coding type of each image block is input into the mapping of entropy coding type-decoding cost reference amount pre-stored in the database, and the decoding cost reference amount corresponding to the entropy coding type of each image block is obtained by mapping and matching.

[0076] The decoding cost proxy is a normalized, dimensionless metric used to quantify the resource consumption required during the decoding of an image block. It is a standardized value calculated based on a preset mapping relationship to measure the decoding complexity of an image block. It represents the comprehensive impact of the average system resource overhead (including CPU decoding latency, memory usage intensity, and buffer management complexity) caused by a specific entropy coding method (such as CABAC, CAVLC, RLE, and ANS). This value is unitless, obtained from a baseline decoding model, and is set to a floating-point number or positive integer between 0 and 1 (for example, 0.8 for CABAC and 0.4 for CAVLC). Higher values ​​indicate higher decoding cost.

[0077] A set of compression feature reference values ​​is extracted from the database, including a reference value of the number of compressed bytes of a block and a reference value of the number of non-zero coefficients of a residual block.

[0078] The compression features of each image block are compared and analyzed with the compression feature reference value, and weighted coupling processing is performed. The decoding cost reference is introduced for comprehensive analysis to obtain the decoding cost characteristic value of each image block. The decoding cost characteristic value is used to quantify the size of the resources required for decoding the image block. The specific calculation process includes: ; Among them, COST x is the decoding cost characteristic value of the x-th image block, num x is the number of bytes of block compression for the xth image block, vz x is the number of non-zero coefficients of the residual block of the x-th image block, num0 is the reference value of the number of bytes of block compression, vz0 is the reference value of the number of non-zero coefficients of the residual block, γ x is the decoding cost index corresponding to the entropy coding type of the x-th image block, β1 is the weighting factor for the number of block compression bytes, β2 is the weighting factor for the number of non-zero coefficients in the residual block, β3 is the weighting factor for the decoding cost index, x is the image block number, x=1,2,3,...,y, and y is the total number of image blocks.

[0079] It should be noted that since these three factors respectively reflect the bitstream volume, transform domain complexity, and decoding load characteristics of the coding structure, direct superposition will lead to imbalance due to different scales and distribution characteristics. Therefore, corresponding weighting factors are introduced, including a weighting factor for the number of compressed bytes in the block, a weighting factor for the number of non-zero coefficients in the residual block, and a weighting factor for the decoding cost index, to serve as weight regulators for each feature in the overall decoding cost model. The weighting factor for the number of compressed bytes in the block is used to balance the impact of the bitstream volume occupied by the image block on the decoding burden, the weighting factor for the number of non-zero coefficients in the residual block emphasizes the impact of the content complexity of the image block on the computational load, and the weighting factor for the decoding cost index serves as a comprehensive indicator at the coding structure level. The specific acquisition process involves extracting feature parameters corresponding to the image block samples used for model training, including the number of compressed bytes in the block, the number of non-zero coefficients in the residual block, and the decoding cost index. A nonlinear model (such as SVR or XGBoost) is used to construct a prediction model, using the image block features as independent variables and the decoding resource consumption as the dependent variable, and performing fitting and solution. After model training converges, the regression coefficients for each feature item, also known as fitting weights, are extracted. The regression coefficients for each feature item are normalized so that their sum equals 1, resulting in standardized weighting factors. These factors include a weighting factor for the number of compressed bytes in the block, a weighting factor for the number of nonzero coefficients in the residual block, and a weighting factor for the decoding cost reference.

[0080] It's also important to note that there's a correlation between the three parameters: block compression byte count, entropy coding type, and the number of non-zero coefficients in the residual block. Together, they reflect the complexity of the image block during encoding and its potential resource consumption during decoding. They can comprehensively assess the decoding cost of an image block from multiple perspectives, including data volume, structural complexity, and content variability. Block compression byte count directly reflects the bitstream volume occupied by the encoded image block and is influenced by both the coding redundancy compression ratio and the image content complexity. Generally speaking, the richer the image details and the more dramatic the changes, the higher the block compression byte count; however, this value is also closely related to the entropy coding mode used. The entropy coding type determines the structured compression method of the bitstream and is a key factor influencing the mechanism for determining block compression byte count. Complex entropy coding methods such as CABAC can significantly reduce bitstream length (reducing block compression byte count), but they introduce a higher computational burden for context modeling and probability estimation, thereby increasing actual decoding overhead. Therefore, even with similar block compression byte counts, different entropy coding types can still result in significant differences in decoding resource requirements. The number of nonzero coefficients in a residual block measures the complexity of the frequency domain information in the image block in the transform domain (e.g., DCT). More nonzero coefficients indicate a larger prediction residual, meaning the image block contains more detailed information. Consequently, more bits are required to record the transformed values ​​during encoding, increasing the number of bytes required for block compression. Furthermore, the decoding stage requires more frequent inverse quantization and inverse transform calculations, increasing the CPU or GPU load.

[0081] Mapping obtains the required resource status parameters of each image block, including: The decoding cost characteristic value of each image block is input into the mapping set of decoding cost characteristic value-required resource status parameter set pre-stored in the database, and mapping matching is performed to obtain the required resource status parameter set of each image block, including the required CPU load ratio, the required GPU video memory load ratio, the required system memory capacity ratio and the required disk IO resource ratio.

[0082] Real-time hardware resource status parameters are obtained, and based on the candidate decoding sequence and the required resource status parameters of each image block, the number of image blocks that can currently be decoded in parallel is obtained.

[0083] Obtain real-time hardware resource status parameters, and based on the candidate decoding sequence and the required resource status parameters of each image block, obtain the number of image blocks that can currently be decoded in parallel, specifically including: Real-time hardware resource status parameters include the current remaining CPU load ratio, the current remaining GPU memory load ratio, the remaining system memory capacity ratio, and the current remaining disk IO resource ratio.

[0084] In the candidate decoding sequence, they are loaded into the decoding slot one by one in sequence, and their required resource status parameters are accumulated in real time until any real-time hardware resource status parameter is equal to its corresponding accumulated required resource status parameter. The loading is terminated, and the number of image blocks loaded in the decoding slot is the number of image blocks that can currently be decoded in parallel.

[0085] By obtaining real-time hardware resource status parameters and combining the resource demand characteristics of each image block in the candidate decoding sequence, the number of image blocks that can currently be decoded in parallel is dynamically calculated, which can significantly improve the system decoding efficiency and overall stability, achieve a precise match between decoding tasks and system resources, and avoid decoding blockage, frame rate drop or system overload due to unreasonable resource allocation. Make full use of multi-dimensional hardware resources such as CPU, GPU, memory and disk IO at the current moment, maximize the number of concurrent decoding within the scope of resource allowance, thereby improving system throughput and decoding rate. When a certain type of key resource tends to be tight (such as the CPU is close to full load or the video memory usage is too high), the system will automatically terminate the image block loading and dynamically control the degree of concurrency to effectively prevent decoding failures or screen tearing problems caused by resource bottlenecks.

[0086] A judgment is made based on the current number of image blocks that can be decoded in parallel. When the number of image blocks that can be decoded in parallel is less than the target threshold number of image blocks that can be decoded in parallel, a speed-up compensation mechanism is started to perform virtual decoding.

[0087] Based on the required frame rate of the required high-smoothness lossless decoded video, the decoding limit time of each original image frame is obtained, specifically including: the known frame rate refers to the number of image frames displayed per second in the video, and one second is divided according to the frame rate value to obtain the decoding limit time of each original image frame.

[0088] The total number of image blocks of each original image frame is extracted. The total number of image blocks is the number of image blocks that need to be decoded within the decoding limit time.

[0089] The minimum resource status parameter among the real-time hardware resource status parameters is obtained, and the mapping between the corresponding resource status parameter and the number of scheduling cycles in the input database is centrally mapped and matched to obtain the current number of scheduling cycles.

[0090] The current number of scheduling cycles is based on the most bottleneck resource status parameters in real-time hardware resources. It is an integer value or approximate value derived through the mapping relationship between preset resource status parameters and the number of scheduling cycles. It is used to quantify the size of the decoding task batch that the system can support under the current resource constraints. This number serves as the core control parameter in the scheduling mechanism, guiding the system to divide the image blocks in the original image frame into several scheduling batches, and reasonably allocate decoding tasks within each batch according to the resource carrying capacity, thereby achieving dynamic matching of software and hardware resources. By adjusting the number of scheduling cycles, it is possible to optimize resource utilization efficiency, prevent resource overload or idleness, and improve the overall decoding fluency and stability while ensuring that the frame rate requirements are met.

[0091] The decoding limit time is divided into several scheduling cycles according to the number of scheduling cycles, and the number of image blocks that need to be decoded within the decoding limit time is evenly divided into each scheduling cycle to obtain the target parallel decoding image block number threshold of each scheduling cycle.

[0092] The number of parallel decodable image blocks in the decoding slot is monitored in real time. When the number of parallel decodable image blocks is less than the target parallel decodable image block number threshold, it is determined to be a decoding delay and the speed-up compensation mechanism is activated.

[0093] Speed-up compensation mechanism, the specific processing conditions are: The average change intensity value of the image blocks of each original image frame is extracted and compared with the average change intensity threshold preset in the database. If the average change intensity of the image blocks of an original image frame is less than the average change intensity threshold, the original image frame is determined to be a low change intensity image frame.

[0094] If the average change intensity of the image blocks of an original image frame is greater than or equal to the average change intensity threshold and less than the change intensity significance value, the original image frame is determined to be a medium change intensity image frame.

[0095] If the average change intensity of the image blocks of an original image frame is greater than or equal to the change intensity significance value, the original image frame is determined to be a high change intensity image frame.

[0096] The average change intensity difference between the average change intensity value of the low change intensity image frame and the average change intensity threshold is extracted, and the mapping of the average change intensity difference-repair derivation factor pre-stored in the database is input for mapping matching to obtain the repair derivation factor of the low change intensity image frame, and the low change intensity image frame is virtually decoded.

[0097] The significant difference in change intensity between the average change intensity value and the significant change intensity value of the medium change intensity image frame is extracted, and the mapping of the significant change intensity difference value-virtual decoded image block ratio pre-stored in the database is input for centralized mapping matching to obtain the virtual decoded image block ratio of the medium change intensity image frame, and partial virtual decoding is performed on the medium change intensity image frame.

[0098] For image frames with high intensity variation, the normal decoding state is maintained.

[0099] Virtual decoding, specifically including: The image blocks of the previous frame of the low-variation intensity image frame are extracted and corresponded one-to-one with the image blocks of the low-variation intensity image frame to perform image block placeholders. The restoration derivation factors are input into the restoration derivation factor-complete virtual frame parameter mapping set preset in the database for mapping and matching to obtain the complete virtual frame parameters of the low-variation intensity image frame, including pixel interpolation and illumination derivation values. Image restoration is performed on the low-variation intensity image frame to complete the virtual decoding of the low-variation intensity image frame.

[0100] It should be noted that pixel interpolation refers to the use of an appropriate interpolation algorithm (such as bilinear interpolation, cubic convolution interpolation, or an edge-protection-based interpolation method) based on the pixel interpolation parameters obtained from the mapping set. In the embodiment of the present invention, the cubic convolution interpolation method is selected to refine and smooth the mapped pixels, compensate for pixel displacement and detail loss caused by time intervals, and enhance the continuity and visual naturalness of the image blocks. Lighting derivation refers to the dynamic adjustment of the brightness, contrast, and color balance of the image blocks in combination with the lighting derivation value to adapt to changes in scene lighting and avoid lighting inconsistencies caused by directly using the pixels of the previous frame. Specifically, the pixel value distribution of the image block can be adjusted through an illumination correction model (such as adjustment based on the local brightness mean and variance).

[0101] Image blocks of a medium-variable intensity image frame are extracted. Based on the candidate decoding sequence and the ratio of virtual decoded image blocks of the medium-variable intensity image frame, the position coordinates of the virtual decoded image blocks in the medium-variable intensity image frame are obtained. Specifically, a certain number of virtual decoded image blocks are obtained in reverse order of the candidate decoding sequence based on the ratio of virtual decoded image blocks. Since the position coordinates of each image block are included in the candidate decoding sequence, they can be directly retrieved. The image frame before the medium-variable intensity image frame is extracted, and the image blocks with the same position coordinates as the virtual decoded image blocks are extracted. After a one-to-one correspondence, image blocks are occupied, completing partial virtual decoding of the medium-variable intensity image frame.

[0102] In this embodiment, the present invention provides a multi-compatible high-smoothness lossless decoding system, including: The candidate decoding sequence module is used to extract the video with high smoothness and lossless decoding requirements, divide the original image frames in the video with high smoothness and lossless decoding requirements into image blocks, perform topological scheduling on each image block of the original image frame, and generate a candidate decoding sequence.

[0103] The decoding cost calculation module is used to obtain the decoding cost characteristic value of each image block in the candidate decoding sequence and map it to obtain the required resource status parameters of each image block. The required resource status parameters include the required CPU load ratio, the required GPU memory load ratio, the required system memory capacity ratio and the required disk IO resource ratio.

[0104] The parallel decoding module is used to obtain real-time hardware resource status parameters and obtain the number of image blocks that can be decoded in parallel based on the candidate decoding sequence and the required resource status parameters of each image block.

[0105] The virtual decoding module is used to make a judgment based on the current number of image blocks that can be decoded in parallel, and when the number of image blocks that can be decoded in parallel is less than the target number threshold of image blocks that can be decoded in parallel, start the speed-up compensation mechanism to perform virtual decoding.

[0106] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0107] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all details in detail, nor do they limit the invention to specific implementation methods. Obviously, many modifications and changes can be made based on the contents of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can better understand and utilize the present invention. As long as they do not deviate from the structure of the present invention or exceed the scope defined by the present invention, they should fall within the scope of protection of the present invention.

Claims

1. A multi-compatible high-smooth lossless decoding method, characterized in that: include: Extracting a video that requires high-smoothness lossless decoding, dividing the original image frame in the video into image blocks, performing topological scheduling on each image block of the original image frame, and generating a candidate decoding sequence; Obtaining a decoding cost characteristic value for each image block in the candidate decoding sequence, and mapping it to obtain the required resource status parameters for each image block, wherein the required resource status parameters include the required CPU load ratio, the required GPU memory load ratio, the required system memory capacity ratio, and the required disk IO resource ratio; Obtaining real-time hardware resource status parameters, and based on the candidate decoding sequence and the required resource status parameters of each image block, obtaining the number of image blocks that can currently be decoded in parallel; A judgment is made based on the current number of image blocks that can be decoded in parallel. When the number of image blocks that can be decoded in parallel is less than the target threshold number of image blocks that can be decoded in parallel, a speed-up compensation mechanism is started to perform virtual decoding.

2. The multi-compatible, high-smoothness, lossless decoding method according to claim 1, characterized in that: The original image frame in the video requiring high-smoothness lossless decoding is divided into image blocks, and topology scheduling is performed on each image block of the original image frame. The specific process is as follows: Obtaining a required frame rate for a highly smooth lossless decoded video, separating the required highly smooth lossless decoded video into original image frames according to the required frame rate, and dividing the original image frames into image blocks based on a preset fixed granularity; The image block is the basic scheduling unit of the decoding task; The in-degree value of each image block is calculated, and candidate decoding sequences are generated from small to large based on the size of the in-degree value of each image block.

3. The multi-compatible, high-smoothness, lossless decoding method according to claim 2, characterized in that: The specific process of generating a candidate decoding sequence is as follows: When generating candidate decoding sequences, if there are multiple image blocks with the same in-degree value in a certain original image frame, such image blocks are extracted and grouped into the same in-degree image block group. The position coordinates of each image block in the same in-degree image block group are obtained, and the Euclidean distance between each image block and the center point of the original image frame is calculated, which is recorded as the non-centrality. Obtain the temporal change intensity index between each image block in the same in-degree image block group in the original image frame and the image block corresponding to the same position coordinate in the previous image frame, including PSNR gradient change, SSIM gradient change, optical flow intensity, MSE, edge map coincidence and grayscale change. Perform cross-image block mean processing on the temporal change intensity index of each image block in the same in-degree image block group to obtain a mean set of temporal change intensity indexes. This mean set is used as a reference value and compared with the temporal change intensity index of each image block in the same in-degree image block group. Weighted coupling processing is performed to obtain the change intensity value of each image block in the same in-degree image block group. The non-centrality and variation intensity of each image block in the same in-degree image block group are comprehensively fitted to obtain the candidate decoding sequence value of each image block in the same in-degree image block group, and the candidate decoding sequence of each image block in the same in-degree image block group is obtained by sorting them from large to small.

4. The multi-compatible, high-smoothness, lossless decoding method according to claim 1, characterized in that: The obtaining of the decoding cost characteristic value of each image block in the candidate decoding sequence specifically includes: Analyze and obtain the compression characteristics of each image block in the candidate decoding sequence, including the number of block compression bytes, entropy coding type, and the number of non-zero coefficients of the residual block; Input the entropy coding type of each image block into the mapping of entropy coding type-decoding cost reference amount pre-stored in the database, and obtain the decoding cost reference amount corresponding to the entropy coding type of each image block by mapping and matching; Extracting a set of compression feature reference values ​​from a database, including a reference value of the number of compressed bytes of a block and a reference value of the number of non-zero coefficients of a residual block; The compression features of each image block are compared and analyzed with the compression feature reference value, and a weighted coupling process is performed. The decoding cost reference quantity is introduced for comprehensive analysis to obtain the decoding cost characteristic value of each image block. The decoding cost characteristic value is used to quantify the size of the resources required for decoding the image block.

5. The multi-compatible high-smoothness lossless decoding method according to claim 1, characterized in that: The mapping obtains the required resource status parameters of each image block, specifically including: The decoding cost characteristic value of each image block is input into the mapping set of decoding cost characteristic value-required resource status parameter set pre-stored in the database, and mapping matching is performed to obtain the required resource status parameter set of each image block, including the required CPU load ratio, the required GPU video memory load ratio, the required system memory capacity ratio and the required disk IO resource ratio.

6. The multi-compatible, high-smoothness, lossless decoding method according to claim 1, characterized in that: The real-time hardware resource status parameter is obtained, based on the candidate decoding sequence and the required resource status parameter of each image block, to obtain the number of image blocks that can be decoded in parallel, specifically including: The real-time hardware resource status parameters include the current remaining load ratio of the CPU, the current remaining load ratio of the GPU memory, the remaining capacity ratio of the system memory, and the current remaining disk IO resource ratio; In the candidate decoding sequence, they are loaded into the decoding slot one by one in sequence, and their required resource status parameters are accumulated in real time until any real-time hardware resource status parameter is equal to its corresponding accumulated required resource status parameter. The loading is terminated, and the number of image blocks loaded in the decoding slot is the number of image blocks that can currently be decoded in parallel.

7. The multi-compatible, high-smoothness, lossless decoding method according to claim 1, characterized in that: When the number of parallel decodable image blocks is less than the target parallel decodable image block number threshold, starting the speed-up compensation mechanism to perform virtual decoding specifically includes: Based on the required frame rate of the video requiring high-smoothness lossless decoding, the decoding limit time of each original image frame is obtained, the total number of image blocks of each original image frame is extracted, and the number of image blocks that need to be decoded within the decoding limit time is obtained; Obtain the minimum resource status parameter among the real-time hardware resource status parameters, input the corresponding resource status parameter-scheduling cycle number mapping, perform mapping matching, and obtain the current scheduling cycle number; Divide the decoding limit time according to the number of scheduling cycles to obtain a number of scheduling cycles, and evenly divide the number of image blocks that need to be decoded within the decoding limit time into each scheduling cycle to obtain a target parallel decoding image block number threshold for each scheduling cycle; The number of parallel decodable image blocks in the decoding slot is monitored in real time. When the number of parallel decodable image blocks is less than the target parallel decodable image block number threshold, it is determined to be a decoding delay and the speed-up compensation mechanism is activated.

8. The multi-compatible, high-smoothness, lossless decoding method according to claim 1, characterized in that: The specific processing conditions of the speed-up compensation mechanism are as follows: Extract the average change intensity value of the image blocks of each original image frame and compare it with the average change intensity threshold preset in the database. If the average change intensity of the image blocks of an original image frame is less than the average change intensity threshold, the original image frame is determined to be a low change intensity image frame; If the average change intensity of the image blocks of an original image frame is greater than or equal to the average change intensity threshold and less than the change intensity significance value, then the original image frame is determined to be a medium change intensity image frame; If the average change intensity of the image blocks of an original image frame is greater than or equal to the change intensity significance value, the original image frame is determined to be a high change intensity image frame, and the normal decoding state is still maintained for the high change intensity image frame; Extracting the average change intensity difference between the average change intensity value of the low change intensity image frame and the average change intensity threshold, inputting the average change intensity difference-repair derivation factor mapping pre-stored in the database to perform mapping matching to obtain the repair derivation factor of the low change intensity image frame, and performing virtual decoding on the low change intensity image frame; The significant difference in change intensity between the average change intensity value and the significant change intensity value of the medium change intensity image frame is extracted, and the mapping of the significant change intensity difference value-virtual decoded image block ratio pre-stored in the database is input for centralized mapping matching to obtain the virtual decoded image block ratio of the medium change intensity image frame, and partial virtual decoding is performed on the medium change intensity image frame.

9. The multi-compatible, high-smoothness, lossless decoding method according to claim 8, characterized in that: The virtual decoding specifically includes: Extracting each image block of the previous frame of the low-intensity variation image frame, making a one-to-one correspondence with each image block of the low-intensity variation image frame, performing image block placeholders, inputting the restoration derivation factor into a preset restoration derivation factor-complete virtual frame parameter mapping set in the database for mapping and matching, obtaining the complete virtual frame parameters of the low-intensity variation image frame, including pixel interpolation and illumination derivation values, performing image restoration on the low-intensity variation image frame, and completing virtual decoding of the low-intensity variation image frame; An image block of a medium-varying intensity image frame is extracted. Based on the candidate decoding sequence and the ratio of virtual decoded image blocks of the medium-varying intensity image frame, the position coordinates of the virtual decoded image block in the medium-varying intensity image frame are obtained. The previous image frame of the medium-varying intensity image frame is extracted. The image block with the same position coordinates as the virtual decoded image block is extracted. After one-to-one correspondence, the image block is occupied, thereby completing partial virtual decoding of the medium-varying intensity image frame.

10. A system using the multi-compatible high-smoothness lossless decoding method according to any one of claims 1 to 9, characterized in that: The candidate decoding sequence module is used to extract the video with high smoothness and lossless decoding, divide the original image frame in the video with high smoothness and lossless decoding into image blocks, perform topological scheduling on each image block of the original image frame, and generate a candidate decoding sequence; A decoding cost calculation module is used to obtain the decoding cost characteristic value of each image block in the candidate decoding sequence and map it to obtain the required resource status parameters of each image block. The required resource status parameters include the required CPU load ratio, the required GPU memory load ratio, the required system memory capacity ratio, and the required disk IO resource ratio; A parallel decoding module is used to obtain real-time hardware resource status parameters and obtain the number of image blocks that can be decoded in parallel based on the candidate decoding sequence and the required resource status parameters of each image block; The virtual decoding module is used to make a judgment based on the current number of image blocks that can be decoded in parallel, and when the number of image blocks that can be decoded in parallel is less than the target number threshold of image blocks that can be decoded in parallel, start the speed-up compensation mechanism to perform virtual decoding.

Images