Image pre-division method based on multi-layer multi-motion search result

Abstract

The invention relates to an image pre-division method based on a multi-layer multi-motion search result, and belongs to the technical field of video coding. The method comprises the steps that an original image and a prediction search image are acquired, and according to the original image and the prediction search image, motion information of the original image is obtained through multi-layer multi-time motion search; obtaining a video sequence and extracting to obtain an image frame set, coding and extracting features of the image frame set to obtain an image pre-division data set; training and testing according to the image pre-division data set through a decision tree algorithm to obtain an image pre-division decision tree; and obtaining a current image frame, extracting current image motion information through the multi-layer multi-time motion search, and obtaining a current image pre-division state through the image pre-division decision tree according to the current image motion information, thereby assisting an encoder in encoding, accelerating the process of determining the encoding block division form, and improving the encoding efficiency. The calculation consumption of the encoder is reduced, and efficient and accurate image pre-division is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of video encoding, and in particular relates to an image pre-division method based on multi-layer and multi-motion search results. Background Art

[0002] In the field of image and video, especially in video surveillance, users are often more concerned about moving objects and people, and pay less attention to the unchanging trees, roads and other areas in the background. Therefore, when compressing and encoding videos, moving objects require higher clarity, and background objects require a certain degree of stability. In response to the different needs of the two, the current mainstream encoding standards also have common characteristics in the encoding results: the encoding blocks in the moving area are smaller, the residual value of the compression result is larger, and it contains more detailed features; in contrast, the encoding blocks in the static area are larger, and the residual value of the compression result is smaller, so as to ensure that the different frames of the entire sequence are kept as consistent as possible.

[0003] Motion search is a common method in current mainstream coding standards. By detecting the position of the pixel points corresponding to the moving object in the video, the change of the moving pixel in the time domain and its correlation between adjacent frames can be obtained. Based on the corresponding relationship between the previous frame and the current frame, the instantaneous speed and direction of movement of the moving object can be calculated relatively easily. Mainstream coding standards require motion search between the reconstructed image and the current image, calculate the residual of the image block and then encode it to improve the compression rate of the final bitstream.

[0004] In H.264, the size of the coding block is fixed, while in H.265, a coding block can be divided into multiple small coding blocks in the form of a quadtree. On the one hand, large coding blocks can improve the coding efficiency of the background area, and on the other hand, small coding blocks can better process local details of the image, making the prediction of complex areas more accurate. In H.266, a coding block can be divided into more forms. In addition to the quadtree, it can also be divided into a binary tree and a ternary tree. The rich coding block format has played a powerful role in further improving the coding efficiency and improving the clarity of the reconstructed image, but it also brings a great burden to the encoder during processing. Taking H.265 as an example, when encoding using a quadtree-based cyclic hierarchical structure, a basic coding unit (CTU) may have multiple division forms. Traversing all CU blocks in the entire CTU to find the best CTU division structure will bring a great computational burden to the encoder. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention proposes an image pre-division method based on multi-layer and multiple motion search results. By analyzing the motion information, the segmentation form of the coding block is determined in advance, and the number of loop iteration layers of the coding block is reduced to reduce the coding consumption.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An image pre-division method based on multi-layer and multi-motion search results, comprising:

[0008] Acquire an original image and a predicted search image, and obtain the original image motion information through multiple layers and multiple motion searches according to the original image and the predicted search image, wherein the predicted search image includes a reconstructed image of the current image and an image frame before the current image;

[0009] Acquire a video sequence and extract an image frame set, encode and feature extract the image frame set to obtain an image pre-divided data set;

[0010] According to the image pre-division data set, an image pre-division decision tree is obtained by training and testing a decision tree algorithm;

[0011] The current image frame is acquired and the current image motion information is obtained through the multi-layer and multi-motion search extraction. The current image pre-division state is obtained through the image pre-division decision tree according to the current image motion information. The encoder is assisted in encoding according to the current image pre-division state.

[0012] Preferably, obtaining the original image motion information through multi-layer and multi-motion searches according to the original image and the predicted search image comprises:

[0013] The first motion information of the original image is obtained by performing a hierarchical precision motion search based on the original image and the reconstructed image of the current image;

[0014] The second motion information of the original image is obtained by performing a hierarchical precision motion search based on the original image and the previous frame of the current image;

[0015] The original image motion information is obtained by integrating the first motion information of the original image and the second motion information of the original image.

[0016] Preferably, the hierarchical precision motion search comprises:

[0017] Obtaining sub-precision search motion information through multi-layer downsampling according to the original image and the predicted search image, wherein the sub-precision search motion information includes integer pixel size search motion information, 2×2 mean downsampling search motion information, and 4×4 mean downsampling search motion information;

[0018] The multi-layer downsampling includes obtaining an integer pixel size search motion information through a hierarchical search based on the original image and the predicted search image, and the hierarchical search is to obtain the motion information by searching the sub-layer CU block through full pixel matching; obtaining a 2×2 mean downsampled image by 2×2 mean downsampling the original image, and obtaining the 2×2 mean downsampled search motion information through a hierarchical search based on the 2×2 mean downsampled image; obtaining a 4×4 mean downsampled image by 4×4 mean downsampling the original image, and obtaining the 4×4 mean downsampled search motion information through a hierarchical search based on the 4×4 mean downsampled image.

[0019] Preferably, the hierarchical search includes:

[0020] Obtain an original image coding unit CTU, and divide the original image coding unit CTU into sub-layer CU blocks;

[0021] Preset best matching block search range:

[0022] Obtaining the best matching block of the current CU block in the prediction search image by block matching according to the best matching block search range;

[0023] The motion information is recorded and integrated according to the best matching block.

[0024] Preferably, the step of obtaining an image pre-division decision tree by training and testing the image pre-division data set through a decision tree algorithm comprises:

[0025] According to the image pre-divided data set, the image pre-divided training set and the image pre-divided test set are obtained by data set division;

[0026] Generate an initial image pre-division decision tree through the CART algorithm according to the image pre-division training set;

[0027] The image pre-division decision tree is obtained by evaluating the initial image pre-division decision tree through static testing according to the image pre-division test set.

[0028] Preferably, generating an initial image pre-division decision tree by using a CART algorithm according to the image pre-division training set comprises:

[0029] Initialize the decision tree and determine the root node of the decision tree;

[0030] Preset the maximum number of layers of the decision tree;

[0031] The decision tree is trained by the CART algorithm using the image motion information in the image pre-division training set as a feature, wherein the image motion information includes but is not limited to: the best motion vector of the current CU block, the Sad of the current CU block, the Cost of the current CU block, the residual under the best motion vector of the current CU block, the variance of the residual under the best motion vector of the current CU block, the inner product of the residual under the best motion vector of the current CU block, the best motion vector of the sub-layer CU block, the Sad of the sub-layer CU block, the Cost of the sub-layer CU block, the difference value of multiple motion vectors of the sub-layer CU block, and the difference value of multiple Costs of the sub-layer CU block;

[0032] The decision tree is optimized by post-pruning to obtain an initial image pre-segmentation decision tree.

[0033] Preferably, evaluating the initial image pre-division decision tree through static testing according to the image pre-division test set to obtain the image pre-division decision tree comprises:

[0034] Obtaining an image pre-division state through the initial image pre-division decision tree according to the image pre-division test set, wherein the image pre-division state includes a confidence level and a pre-division result;

[0035] For a CU block with a higher confidence level, the encoder uses the pre-division result of the initial image pre-division decision tree to encode to obtain a predictive coding image sequence;

[0036] An image pre-division decision tree is obtained according to the predicted coded image sequence and the real coded image sequence through objective performance PSNR evaluation.

[0037] An image pre-division system based on multi-layer and multi-motion search results, the system is applied to the above-mentioned image pre-division method based on multi-layer and multi-motion search results, including a multi-layer and multi-motion search module, an image pre-division decision tree module, and an image pre-division module;

[0038] The multi-layer and multi-motion search module is used to obtain the original image and the predicted search image and obtain the original image motion information through multi-layer and multi-motion search;

[0039] The image pre-division decision tree module is used to obtain a video sequence and extract an image frame set, encode and feature extract the image frame set to obtain an image pre-division data set, and obtain an image pre-division decision tree through decision tree algorithm training and testing;

[0040] The image pre-division module is used to obtain the current image frame, obtain the current image motion information through the multi-layer and multiple motion search extraction, obtain the current image pre-division state through the image pre-division decision tree, and assist the encoder to encode according to the current image pre-division state.

[0041] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the image pre-division method based on multi-layer and multi-motion search results is implemented.

[0042] A storage medium containing computer executable instructions, wherein the computer executable instructions are used to execute the above-mentioned image pre-division method based on multi-layer and multi-motion search results when executed by a computer processor.

[0043] The beneficial effects of the present invention are:

[0044] (1) Comprehensive motion information is obtained through multi-layer and multiple motion searches, and the pre-division state of the CU is obtained through the image pre-division decision tree, thereby assisting the encoder in final encoding. The CU block pre-division result with high confidence is directly used for encoding, and no internal CU iteration is performed, which reduces the workload of the encoder and improves the encoding efficiency. Compared with the general algorithm that determines the CU segmentation result by obtaining the rate-distortion cost of the prediction mode, it greatly reduces the difficulty of feature acquisition, speeds up the process of determining the segmentation form of the coding block, effectively saves the time of recursive calculation of the rate-distortion rate in the CTU loop during encoding, and reduces the computational consumption of the encoder.

[0045] (2) By presetting the maximum number of layers of the decision tree, we can prevent the tree from being too complex while maintaining the generalization ability of the model. By post-pruning the decision tree, we can trim redundant features, remove nodes that do not significantly improve model performance, and reduce the complexity of the model to prevent overfitting.

[0046] (3) By predicting only fixed-shape CUs, the complexity of prediction during encoding can be greatly reduced, and the number of cycles of each CU in the CTU can be reduced, which greatly saves computational complexity and encoding time. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0048] Figure 1 The figure is a flow chart of an image pre-division method based on multi-layer and multi-motion search results of the present invention. DETAILED DESCRIPTION

[0049] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0050] See also Figure 1 , an image pre-segmentation method based on multi-layer and multi-motion search results, comprising:

[0051] S1: Acquire an original image and a predicted search image, and obtain the original image motion information through multiple layers and multiple motion searches according to the original image and the predicted search image, wherein the predicted search image includes a reconstructed image of the current image and an image frame before the current image;

[0052] S2: Acquire a video sequence and extract an image frame set, encode and feature extract the image frame set to obtain an image pre-divided data set;

[0053] S3: training and testing the image pre-division data set through a decision tree algorithm to obtain an image pre-division decision tree;

[0054] S4: Acquire the current image frame and extract the current image motion information through the multi-layer and multiple motion search, obtain the current image pre-division state through the image pre-division decision tree according to the current image motion information, and assist the encoder to encode according to the current image pre-division state.

[0055] In this embodiment, obtaining the original image motion information through multi-layer and multi-motion search according to the original image and the predicted search image is specifically implemented by the following steps:

[0056] S101: reconstructing an image based on an original image and a current image to obtain first motion information of the original image through a hierarchical precision motion search;

[0057] S102: obtaining second motion information of the original image through hierarchical precision motion search based on the original image and the previous frame of the current image;

[0058] The hierarchical precision motion search is specifically implemented by the following steps:

[0059] Obtaining sub-precision search motion information through multi-layer downsampling according to the original image and the predicted search image, wherein the sub-precision search motion information includes integer pixel size search motion information, 2×2 mean downsampling search motion information, and 4×4 mean downsampling search motion information;

[0060] Specifically, the motion information is searched for int-pixel size by hierarchical search according to the original image and the predicted search image, wherein the hierarchical search is to obtain the motion information by searching the sub-layer CU block by full pixel matching;

[0061] Downsampling the original image by 2×2 mean to obtain a 2×2 mean downsampled image;

[0062] Obtaining 2×2 mean down-sampling search motion information through a hierarchical search according to the 2×2 mean down-sampling image;

[0063] Obtaining a 4×4 mean downsampled image by 4×4 mean downsampling the original image;

[0064] The 4×4 mean down-sampling search motion information is obtained through a hierarchical search according to the 4×4 mean down-sampling image.

[0065] The hierarchical search is specifically implemented by the following steps:

[0066] Obtain an original image coding unit CTU, and divide the original image coding unit CTU into sub-layer CU blocks;

[0067] Preset best matching block search range:

[0068] Obtaining the best matching block of the current CU block in the prediction search image by block matching according to the best matching block search range;

[0069] The motion information is recorded and integrated according to the best matching block.

[0070] The block matching is specifically implemented by the following steps:

[0071] For the candidate block at the corresponding position in the predicted search image, a matching cost is obtained by cost calculation, and the mathematical expression of the matching cost is:

[0072] Cost=Sad+Lambda×abs(MV-MVP),

[0073] Among them, Cost is the matching cost, Sad is the absolute value difference of each pixel value in the CU, Lambda is a fixed threshold, MV is the motion information of the matching process, and MVP is the motion information of the neighboring block of the current block.

[0074] The candidate block at the corresponding position in the prediction search image with the smallest matching cost is selected as the best matching block of the current CU block in the prediction search image.

[0075] It should be noted that the reconstructed image is an image reconstructed based on the coding information.

[0076] S103: Obtain the original image motion information by integrating the original image first motion information and the original image second motion information.

[0077] In this embodiment, the acquisition of a video sequence and extraction to obtain an image frame set, encoding and feature extraction of the image frame set to obtain an image pre-divided data set are specifically implemented by the following steps:

[0078] S201: Obtain CU segmentation information of each layer of each image by encoding the image frame set;

[0079] S202: extracting CU motion information of each layer of each image through the multi-layer and multi-layer motion search according to the image frame set;

[0080] S203: Obtain an image pre-divided data set according to the CU motion information of each layer of each image and the CU segmentation information of each layer of each image.

[0081] In this embodiment, the image pre-division decision tree is obtained by training and testing the image pre-division data set through a decision tree algorithm, specifically by the following steps:

[0082] S301: obtaining an image pre-divided training set and an image pre-divided test set by dividing the image pre-divided data set;

[0083] S302: Generate an initial image pre-division decision tree by using a CART algorithm according to the image pre-division training set;

[0084] S302-1: Initialize the decision tree and determine the root node of the decision tree;

[0085] Specifically, the initialized decision tree contains only one undivided node, which represents the image sample in the image pre-divided training set.

[0086] S302-2: Preset the maximum number of layers of the decision tree to prevent the tree from being too complex while maintaining the generalization ability of the model;

[0087] S302-3: Using the image motion information in the image pre-division training set as a feature, the decision tree is trained by the CART algorithm, wherein the image motion information includes but is not limited to: the best motion vector of the current CU block, the Sad of the current CU block, the Cost of the current CU block, the residual under the best motion vector of the current CU block, the variance of the residual under the best motion vector of the current CU block, the inner product of the residual under the best motion vector of the current CU block, the best motion vector of the sub-layer CU block, the Sad of the sub-layer CU block, the Cost of the sub-layer CU block, the difference value of multiple motion vectors of the sub-layer CU block, and the difference value of multiple Costs of the sub-layer CU block;

[0088] S302-31: For all samples of the current node, calculate information entropy based on motion information features, where the information entropy reflects the degree of confusion of the data;

[0089] S302-32: traverse all features and possible segmentation points, and select the segmentation point that reduces the information entropy to the greatest extent as the optimal segmentation point to generate a decision tree structure;

[0090] S302-33: Split the current node into two sub-nodes according to the optimal segmentation point, each sub-node containing samples in the image pre-division training set that meet the segmentation condition.

[0091] S302-4: Optimizing the decision tree by post-pruning to obtain an initial image pre-segmentation decision tree;

[0092] The post-pruning is to prune after the decision tree is fully generated, cut redundant features, and remove unnecessary branches to prevent overfitting; the performance of each node is evaluated through cross-validation, and nodes that do not significantly improve the model performance are removed, that is, the features of the decision tree that do not contribute much to the pre-division are removed to reduce the complexity of the model.

[0093] S303: Evaluating the initial image pre-division decision tree through static testing according to the image pre-division test set to obtain an image pre-division decision tree;

[0094] S303-1: Obtaining an image pre-division state through the initial image pre-division decision tree according to the image pre-division test set, wherein the image pre-division state includes a confidence level and a pre-division result;

[0095] S303-2: For a CU block with a higher confidence level, the encoder uses the pre-division result of the initial image pre-division decision tree to encode to obtain a predictive coding image sequence;

[0096] S303-3: Obtaining an image pre-division decision tree through objective performance PSNR evaluation according to the predicted coded image sequence and the real coded image sequence;

[0097] Specifically, when the PSNR performance loss of the predicted coded image sequence and the real coded image sequence is controlled within 5%, an image pre-division decision tree is obtained.

[0098] In this embodiment, the current image frame is acquired and the current image motion information is obtained through the multi-layer and multi-motion search extraction, the current image pre-division state is obtained through the image pre-division decision tree according to the current image motion information, and the encoder is assisted to perform encoding according to the current image pre-division state. Specifically, the following steps are implemented:

[0099] Each layer of CU calculates its own image pre-division decision tree, and statistics the pre-division results and pre-division confidence calculated by the image pre-division decision tree to assist the encoder in final encoding. The corresponding CU block is used for encoding, and internal CU iteration is no longer performed, which reduces the workload of the encoder and improves encoding efficiency.

[0100] In this embodiment, the CU division process takes H.265 as an example. For a CTU of size 64×64, there are various division forms such as CU64 division / non-division, 4 CU32 division / non-division within it, and 16 CU16 division / non-division. The image pre-division decision tree is used to determine whether the current layer of CTU64 is divided. If divided, the CTU is not predicted and directly divided into 4 CU32s; if not divided, the prediction of the sub-layer CU is no longer performed, and the prediction result of CTU64 is directly used as the encoding result of the current CTU. Similarly, for CU32 / 16, each layer directly determines the size of the encoded CU according to the result of the image pre-division decision tree until it is divided into the smallest coding unit.

[0101] An image pre-division system based on multi-layer and multi-motion search results includes a multi-layer and multi-motion search module, an image pre-division decision tree module, and an image pre-division module;

[0102] The multi-layer and multi-motion search module is used to obtain the original image and the predicted search image and obtain the original image motion information through multi-layer and multi-motion search;

[0103] The image pre-division decision tree module is used to obtain a video sequence and extract an image frame set, encode and feature extract the image frame set to obtain an image pre-division data set, and obtain an image pre-division decision tree through decision tree algorithm training and testing;

[0104] The image pre-division module is used to obtain the current image frame, obtain the current image motion information through the multi-layer and multiple motion search extraction, obtain the current image pre-division state through the image pre-division decision tree, and assist the encoder to encode according to the current image pre-division state.

[0105] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0106] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0107] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0108] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An image pre-segmentation method based on multi-layer and multi-motion search results, characterized in that: include: Acquire an original image and a predicted search image, and obtain the original image motion information through multiple layers and multiple motion searches according to the original image and the predicted search image, wherein the predicted search image includes a reconstructed image of the current image and an image frame before the current image; Acquire a video sequence and extract an image frame set, encode and feature extract the image frame set to obtain an image pre-divided data set; According to the image pre-division data set, an image pre-division decision tree is obtained by training and testing a decision tree algorithm; The current image frame is acquired and the current image motion information is obtained through the multi-layer and multi-motion search extraction. The current image pre-division state is obtained through the image pre-division decision tree according to the current image motion information. The encoder is assisted in encoding according to the current image pre-division state.

2. The image pre-segmentation method based on multi-layer and multi-motion search results according to claim 1, characterized in that: The obtaining the original image motion information through multi-layer and multi-motion search according to the original image and the predicted search image comprises: The first motion information of the original image is obtained by performing a hierarchical precision motion search based on the original image and the reconstructed image of the current image; The second motion information of the original image is obtained by performing a hierarchical precision motion search based on the original image and the previous frame of the current image; The original image motion information is obtained by integrating the first motion information of the original image and the second motion information of the original image.

3. The image pre-segmentation method based on multi-layer and multi-motion search results according to claim 2, characterized in that: The hierarchical precision motion search includes: Obtaining sub-precision search motion information through multi-layer downsampling according to the original image and the predicted search image, wherein the sub-precision search motion information includes integer pixel size search motion information, 2×2 mean downsampling search motion information, and 4×4 mean downsampling search motion information; The multi-layer downsampling includes obtaining an integer pixel size search motion information through a hierarchical search based on the original image and the predicted search image, and the hierarchical search is to obtain the motion information by searching the sub-layer CU block through full pixel matching; obtaining a 2×2 mean downsampled image by 2×2 mean downsampling the original image, and obtaining the 2×2 mean downsampled search motion information through a hierarchical search based on the 2×2 mean downsampled image; obtaining a 4×4 mean downsampled image by 4×4 mean downsampling the original image, and obtaining the 4×4 mean downsampled search motion information through a hierarchical search based on the 4×4 mean downsampled image.

4. The image pre-segmentation method based on multi-layer and multi-motion search results according to claim 3, characterized in that: The hierarchical search includes: Obtain an original image coding unit CTU, and divide the original image coding unit CTU into sub-layer CU blocks; Preset best matching block search range: Obtaining the best matching block of the current CU block in the prediction search image by block matching according to the best matching block search range; The motion information is recorded and integrated according to the best matching block.

5. The image pre-segmentation method based on multi-layer and multi-motion search results according to claim 1, characterized in that: The step of obtaining an image pre-division decision tree by training and testing the image pre-division data set through a decision tree algorithm comprises: According to the image pre-divided data set, the image pre-divided training set and the image pre-divided test set are obtained by data set division; Generate an initial image pre-division decision tree through the CART algorithm according to the image pre-division training set; The image pre-division decision tree is obtained by evaluating the initial image pre-division decision tree through static testing according to the image pre-division test set.

6. The image pre-segmentation method based on multi-layer and multi-motion search results according to claim 5, characterized in that: Generating an initial image pre-division decision tree by using a CART algorithm according to the image pre-division training set comprises: Initialize the decision tree and determine the root node of the decision tree; Preset the maximum number of layers of the decision tree; The decision tree is trained by the CART algorithm using the image motion information in the image pre-division training set as a feature, wherein the image motion information includes but is not limited to: the best motion vector of the current CU block, the Sad of the current CU block, the Cost of the current CU block, the residual under the best motion vector of the current CU block, the variance of the residual under the best motion vector of the current CU block, the inner product of the residual under the best motion vector of the current CU block, the best motion vector of the sub-layer CU block, the Sad of the sub-layer CU block, the Cost of the sub-layer CU block, the difference value of multiple motion vectors of the sub-layer CU block, and the difference value of multiple Costs of the sub-layer CU block; The decision tree is optimized by post-pruning to obtain an initial image pre-segmentation decision tree.

7. The image pre-segmentation method based on multi-layer and multi-motion search results according to claim 5, characterized in that: The step of evaluating the initial image pre-division decision tree through static testing according to the image pre-division test set to obtain the image pre-division decision tree comprises: Obtaining an image pre-division state through the initial image pre-division decision tree according to the image pre-division test set, wherein the image pre-division state includes a confidence level and a pre-division result; For a CU block with a higher confidence level, the encoder uses the pre-division result of the initial image pre-division decision tree to encode to obtain a predictive coding image sequence; An image pre-division decision tree is obtained according to the predicted coded image sequence and the real coded image sequence through objective performance PSNR evaluation.

8. An image pre-division system based on multi-layer and multi-motion search results, characterized in that: The system is applied to the image pre-division method based on multi-layer and multi-motion search results as described in any one of claims 1 to 7, comprising a multi-layer and multi-motion search module, an image pre-division decision tree module, and an image pre-division module; The multi-layer and multi-motion search module is used to obtain the original image and the predicted search image and obtain the original image motion information through multi-layer and multi-motion search; The image pre-division decision tree module is used to obtain a video sequence and extract an image frame set, encode and feature extract the image frame set to obtain an image pre-division data set, and obtain an image pre-division decision tree through decision tree algorithm training and testing; The image pre-division module is used to obtain the current image frame, obtain the current image motion information through the multi-layer and multiple motion search extraction, obtain the current image pre-division state through the image pre-division decision tree, and assist the encoder to encode according to the current image pre-division state.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the image pre-division method based on multi-layer and multi-motion search results as described in any one of claims 1-7 is implemented.

10. A storage medium containing computer executable instructions, characterized in that: When the computer executable instructions are executed by a computer processor, they are used to perform the image pre-division method based on multi-layer and multi-motion search results as described in any one of claims 1-7.

Images