CN-122002036-A - Intelligent image compression system and method based on multi-mode fusion

CN122002036ACN 122002036 ACN122002036 ACN 122002036ACN-122002036-A

Abstract

The invention relates to the technical field of image processing and discloses an intelligent image compression system and method based on multi-mode fusion, wherein the intelligent image compression system and method comprises an intelligent pre-analysis layer, a core processing layer and a fusion optimization layer, wherein the intelligent pre-analysis layer is used for predicting and outputting an optimal compression scheme for balancing image quality and storage efficiency through a pre-trained self-adaptive decision tree model based on multi-dimensional feature analysis results, the core processing layer is used for calling a format conversion module and a color quantization module to perform standardized conversion and color gamut optimization processing on an image, then performing intelligent image compression without visual damage and image limit compression based on rate distortion optimization respectively, and the fusion optimization layer is used for automatically combining an intelligent pre-analysis module, the color quantization module, an intelligent compression module and a limit compression module and forming an optimal image compression path. The invention can realize high compression rate by a multi-mode fusion technology, is greatly superior to the compression level of a single technology, and ensures the visual lossless image quality effect by a perception optimization algorithm and a self-adaptive selection strategy.

Inventors

JIANG LAN
WU XIAODAN
CAO HUA
CHEN YU
FENG YUSHEN

Assignees

福州大学附属省立医院

Dates

Publication Date: 20260508
Application Date: 20260228

Claims (10)

1. An intelligent image compression system based on multi-modal fusion, the system comprising: The unified management layer is used for receiving user instructions and input images through a unified user interface and providing configuration management functions including application programming interface key management, module selection and parameter configuration; The intelligent pre-analysis layer is used for executing multi-dimensional feature analysis on the input image, and based on a multi-dimensional feature analysis result, the optimal compression scheme for balancing the image quality and the storage efficiency is predicted and output through a pre-trained self-adaptive decision tree model; The core processing layer is used for calling the format conversion module and the color quantization module to perform standardized conversion and color gamut optimization processing on the image according to an optimal compression scheme output by the intelligent pre-analysis layer, then driving the intelligent compression module and the limit compression module to work in parallel, and respectively executing intelligent image compression without visual damage and image limit compression based on rate distortion optimization; The fusion optimization layer is used for automatically combining the intelligent pre-analysis module, the color quantization module, the intelligent compression module and the limit compression module according to an optimal compression scheme output by the intelligent pre-analysis layer and a plurality of intermediate results generated by parallel processing of the core processing layer, forming an optimal image compression path and outputting an optimal compressed image according to the optimal image compression path.
2. The intelligent image compression system based on multi-modal fusion according to claim 1, wherein the intelligent pre-analysis layer, when performing multi-dimensional feature analysis on the input image, predicts and outputs an optimal compression scheme balancing image quality and storage efficiency through a pre-trained adaptive decision tree model based on the multi-dimensional feature analysis result, comprises: extracting multi-dimensional features containing edge density, texture complexity and color richness from an input image, and constructing feature vectors based on the multi-dimensional features; Acquiring historical compression data, sequentially performing feature normalization, outlier processing and data balancing operation on the historical compression data, and constructing and training a self-adaptive decision tree model based on multi-feature fusion and Bayesian optimization according to the preprocessed historical compression data; And inputting the feature vector into a self-adaptive decision tree model, predicting the optimal compression parameter by utilizing the self-adaptive decision tree model and combining a preset decision rule, and dynamically generating an optimal compression strategy by combining the optimal compression parameter and the image type.
3. The intelligent image compression system based on multi-modal fusion according to claim 1, wherein the core processing layer comprises: The format conversion module is used for uniformly converting the input images with various formats into a standard PNG format, and meanwhile, retaining DPI metadata and transparency information; The color quantization module is used for carrying out color quantization on the standard image subjected to format conversion by utilizing an adaptive multi-scale perception clustering algorithm so as to adaptively reserve visual detail characteristics while reducing the number of the image colors; the intelligent compression module is used for calling a cloud intelligent algorithm and combining a perception optimization technology to execute intelligent image compression on the premise of maintaining visual quality; And the limit compression module is used for executing image limit compression by utilizing a binary search and rate distortion optimization strategy and controlling the image compression quality by combining the structural similarity index and the self-adaptive threshold.
4. A multimodal fusion-based intelligent image compression system according to claim 3 wherein the color quantization module, when using an adaptive multiscale perceptual clustering algorithm to color quantize the format-converted standard image to adaptively preserve visual detail features while reducing the number of image colors, comprises: calculating the global edge density and visual saliency of the standard image subjected to format conversion, and dynamically distributing the weight coefficient of each distance component based on the calculation result; Respectively calculating the local weighted color distance, the global color histogram distance and the neighborhood context distance of the color clusters in the standard image, and carrying out weighted fusion according to the weight coefficient of each distance component to obtain the comprehensive perception distance; Simultaneously evaluating color loss, space consistency loss and frequency domain loss according to the comprehensive perception distance, and determining the merging priority of the color clusters by taking the perception loss constraint as a criterion; Dividing the standard image into a quadtree structure, traversing and merging only nodes adjacent to the space position in the quadtree structure in the merging operation of the color clusters, and outputting the image after the color quantization after the merging is completed.
5. The intelligent image compression system based on multi-modal fusion according to claim 4, wherein the dividing the standard image into the quadtree structure, in the merging operation of the color clusters, only traversing and merging the nodes adjacent to the spatial position in the quadtree structure, and outputting the quantized image after the merging is completed comprises: Recursively dividing the standard image into a quadtree structure, wherein each node contains one color cluster, leaf nodes correspond to single pixels or initial color areas, and initializing statistical information of the color clusters for each node; For each node of the current layer, calculating the multiscale sensing distance between the node and the adjacent node on the space position, and comprehensively evaluating the color loss, the space consistency loss and the frequency domain loss of the current node pair; If the total perceived loss of the current node pair is lower than a preset threshold, merging the current node pair into a new father node, inheriting the pixel set of the atomic node by the new father node, updating the statistical information of the color cluster, and after each merging operation, recalculating the dynamic weight coefficient of the new father node according to the pixel number of the two nodes involved in merging; And repeatedly executing the traversing, evaluating and merging processes of adjacent nodes in the next round based on the dynamic weight coefficient of the new father node until the number of the nodes reaches the preset target color number, the total perceived loss of all the mergeable node pairs exceeds a preset threshold or the image compression rate meets the requirement, and outputting the color quantized image based on the color clusters corresponding to the leaf nodes in the quadtree structure.
6. The intelligent image compression system based on multi-modal fusion according to claim 5, wherein the intelligent compression module, when used for invoking a cloud intelligent algorithm and combining with a perception optimization technique, performs intelligent image compression on the premise of maintaining visual quality, comprises: Sequentially carrying out format standardization, size inspection and color space conversion treatment on an input image to obtain a preprocessed image, respectively calculating definition, noise level and complexity of the preprocessed image and constructing a quality feature vector; In the cloud API call preparation stage, checking a service quota state according to the verified API key, and if the quota is insufficient, starting a standby compression scheme, and if the quota is sufficient, encoding the preprocessed image into a byte stream in a specified format; The method comprises the steps of sending a compression request to a cloud terminal, judging whether the compression request is accepted or not by the cloud terminal according to an image compression level and a quality target threshold value, downloading a compressed image if the request is successful, and calculating a structural similarity index and a peak signal-to-noise ratio between the compressed image and an original image; Judging whether the quality of the compressed image is qualified or not by combining the structural similarity index and the peak signal-to-noise ratio, if the quality of the compressed image is qualified, performing post-processing optimization on the compressed image, outputting a final compressed image, and if the quality of the compressed image is unqualified, starting a standby compression scheme; The standby compression scheme is processed by calling a local compression algorithm and adopting a progressive quality parameter reduction mode until the structural similarity index and the peak signal-to-noise ratio of the output image reach a preset minimum acceptable quality threshold.
7. The intelligent image compression system based on multi-modal fusion of claim 6, wherein the limit compression module when configured to perform image limit compression using a binary search and rate distortion optimization strategy and to control image compression quality in combination with structural similarity indicators and adaptive thresholds comprises: calculating a basic threshold value based on texture complexity and a quality target threshold value of an input image, carrying out gain correction by combining an image compression level indicated by the basic threshold value, and then adjusting according to a preset compression mode preference to dynamically generate a structural similarity index threshold value; initializing and calculating by using a golden section algorithm to obtain two golden section points, compressing an image by taking the two golden section points as quality parameters, and synchronously calculating the file size of the compression result and the corresponding structural similarity index value; Comparing the two compression results according to a rate distortion optimization criterion of a result with smaller file under the condition that the structural similarity index threshold is met, updating a search interval based on the comparison result and calculating a next group of golden section points; And repeatedly executing the processes of compressing, evaluating and updating the search interval by two points based on the next group of golden partition points until convergence conditions are met, and selecting the version with the smallest file from all compression results meeting the structural similarity index threshold as the optimal compression result to output.
8. The intelligent image compression system based on multi-mode fusion according to claim 1, wherein the fusion optimization layer automatically combines the intelligent pre-analysis module, the color quantization module, the intelligent compression module and the limit compression module to form an optimal image compression path according to an optimal compression scheme output by the intelligent pre-analysis layer and a plurality of intermediate results generated by parallel processing of the core processing layer, and when outputting an optimal compressed image according to the optimal image compression path, the fusion optimization layer comprises: Receiving an optimal compression scheme which is output by the intelligent pre-analysis layer and contains an image compression path identifier, and calling and combining processing logic in the intelligent pre-analysis module, the color quantization module, the intelligent compression module and the limit compression module according to the compression path identifier so as to execute a corresponding image compression path; The image compression comprises a first compression path aiming at a color simple image, a second compression path aiming at an edge complex image and a third compression path aiming at a comprehensive scene; and fusing image compression results output by the first compression path, the second compression path and the third compression path to obtain an optimal compression image meeting an optimal compression scheme.
9. The intelligent image compression system based on multi-modal fusion according to claim 8, wherein the invoking and combining the processing logic in the intelligent pre-analysis module, the color quantization module, the intelligent compression module, and the limit compression module according to the image compression path identification to execute the corresponding image compression path comprises: For a first compression path of the color simple image, driving a color quantization module to perform color quantization processing on the input image, calculating the image compression rate after color quantization, directly outputting an optimized lossless image if the compression rate meets a preset threshold, otherwise, continuously driving a limit compression module to perform image limit compression processing, and outputting a high-quality lossy image; Aiming at a second compression path of the edge complex image, driving the intelligent compression module to carry out intelligent compression processing on the input image and then carrying out structural similarity index verification, outputting a high-fidelity diagnosis image if the verification is passed, and driving the limit compression module to execute image limit compression processing and then outputting a minimum image reaching the standard if the verification is not passed; And aiming at a third compression path of the comprehensive scene, sequentially driving the color quantization module and the intelligent compression module to execute color quantization and intelligent compression processing, judging whether comprehensive requirements are met after the intelligent compression processing is completed, directly outputting a balanced optimized image if the comprehensive requirements are met, otherwise, driving the limit compression module to further execute image limit compression processing, and outputting a global optimal image.
10. An intelligent image compression method based on multi-mode fusion, which adopts the intelligent image compression system based on multi-mode fusion as claimed in any one of claims 1-9, and is characterized in that the method comprises the following steps: receiving user instructions and input images through a unified user interface, and providing configuration management functions including application programming interface key management, module selection and parameter configuration; performing multi-dimensional feature analysis on the input image, and predicting and outputting an optimal compression scheme balancing image quality and storage efficiency through a pre-trained self-adaptive decision tree model based on the multi-dimensional feature analysis result; performing standardized conversion and color gamut optimization processing on the image according to an optimal compression scheme, and then respectively executing intelligent image compression without visual damage and image limit compression based on rate distortion optimization according to strategy instructions of the optimal compression scheme; And generating and executing an optimal image compression path according to an optimal compression scheme, a standardized conversion, a color gamut optimization process, an image intelligent compression and an intermediate result generated by image limit compression, and outputting an optimal compressed image.

Description

Intelligent image compression system and method based on multi-mode fusion Technical Field The invention relates to the technical field of image processing, in particular to an intelligent image compression system and method based on multi-mode fusion. Background The existing image compression technology generally faces the core contradiction that the compression efficiency and the visual quality are difficult to be compatible, a single algorithm such as JPEG lossy compression can obtain a higher compression ratio but can introduce obvious image quality loss, while the lossless compression technologies such as PNG and the like can keep the image quality intact, the compression rate is extremely limited, and the application scene with strict requirements on the file volume is difficult to be met generally between 50% and 70%. In addition, although special techniques such as color quantization and perceptually optimized compression can be improved in specific dimensions, the balance problem cannot be fundamentally solved either due to the limited number of colors or due to excessive complexity of algorithms. Currently, there is a lack of intelligent compression schemes on the market that can adaptively select strategies based on image content characteristics. Users often need to manually adjust professional parameters such as quality level, palette, etc., the use threshold is high and the efficiency is low. Meanwhile, various formats (such as PNG, JPG, TIF) are often processed by different tools, and the process is complicated. More prominently, most compression tools discard key metadata such as DPI, transparency channels and the like in the processing process, so that compressed images are difficult to meet the field requirements of severe requirements on information integrity and quality standards such as professional publishing, academic display and the like. In summary, although advanced technologies such as cloud intelligent compression exist, the cloud intelligent compression relies on networks and APIs, and has insufficient applicability in offline, high-security or real-time scenarios. Therefore, the prior art has obvious short plates, and lacks a comprehensive solution which can uniformly support multi-format intelligent self-adaptive compression and remarkably improve the compression rate on the premise of keeping the image quality and metadata to the maximum extent. For the problems in the related art, no effective solution has been proposed at present. Disclosure of Invention Aiming at the problems in the related art, the invention provides an intelligent image compression system and method based on multi-mode fusion, so as to overcome the technical problems in the prior art. For this purpose, the invention adopts the following specific technical scheme: According to one aspect of the present invention, there is provided an intelligent image compression system based on multi-modal fusion, the system comprising: The unified management layer is used for receiving user instructions and input images through a unified user interface and providing configuration management functions including application programming interface key management, module selection and parameter configuration; The intelligent pre-analysis layer is used for executing multi-dimensional feature analysis on the input image, and based on a multi-dimensional feature analysis result, the optimal compression scheme for balancing the image quality and the storage efficiency is predicted and output through a pre-trained self-adaptive decision tree model; The core processing layer is used for calling the format conversion module and the color quantization module to perform standardized conversion and color gamut optimization processing on the image according to an optimal compression scheme output by the intelligent pre-analysis layer, then driving the intelligent compression module and the limit compression module to work in parallel, and respectively executing intelligent image compression without visual damage and image limit compression based on rate distortion optimization; The fusion optimization layer is used for automatically combining the intelligent pre-analysis module, the color quantization module, the intelligent compression module and the limit compression module according to an optimal compression scheme output by the intelligent pre-analysis layer and a plurality of intermediate results generated by parallel processing of the core processing layer, forming an optimal image compression path and outputting an optimal compressed image according to the optimal image compression path. Preferably, the intelligent pre-analysis layer, when executing multi-dimensional feature analysis on the input image, based on the multi-dimensional feature analysis result, predicts and outputs an optimal compression scheme balancing the image quality and the storage efficiency through a pre-trained adaptive decision tree model, and comprises: e