An apparatus for performing rate-distortion optimized JPEG image compression
By generating fixed-bitrate JPEG image compression parameters using support vector machines and combining discrete cosine transform and Huffman entropy coding, the problem of balancing quality loss and compression rate in image transmission is solved, achieving efficient image compression.
Patent Information
- Application Number
- CN202210066825.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-20
- Publication Date
- 2026-01-02
- Estimated Expiration
- 2042-01-20
AI Technical Summary
In existing technologies, image transmission cannot balance image quality loss and image compression rate, resulting in insufficient utilization of channel bandwidth.
A well-trained support vector machine is used to generate fixed bitrate JPEG format image compression parameters through a gray-level co-occurrence matrix feature generation module, a support vector machine module, a compression parameter category encoding and storage module, and an encoding comparison module. Image compression is then performed by combining discrete cosine transform and Huffman entropy coding.
It achieves the goal of minimizing image quality loss while ensuring image transmission efficiency, and meeting the limitations of channel bandwidth and processing resources.
Smart Images
Figure CN116527941B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of image processing, and particularly relates to a device for fixed code rate JPEG format image compression. BACKGROUND
[0002] At present, with the development of imaging systems, image data collected by imaging detectors based on various systems such as photoelectricity and radio frequency is increasingly rich, and imaging data can be transmitted back to the ground station through a data link, and users can formulate different schemes according to different real-time images, which is of great significance for accurate detection and identification of targets. However, due to the large amount of imaging data, the imaging data cannot be directly transmitted to the ground, and needs to be transmitted through a channel after an image compression process. However, when transmitting images in the traditional way, the greater the image compression rate, the more serious the image quality loss, and it is difficult to balance the image quality loss and the image compression rate and fully utilize the channel bandwidth. SUMMARY
[0003] In view of the above analysis, the present application aims to disclose a device for fixed code rate JPEG format image compression, which solves the technical problem that the image transmission in the prior art cannot balance the image quality loss and the image compression rate.
[0004] The present application discloses a device for fixed code rate JPEG format image compression, which comprises an image acquisition module, an image compression parameter generation module and a JPEG compression module.
[0005] The image acquisition module is used to load the compressed image to be compressed for fixed code rate JPEG format compression.
[0006] The image compression parameter generation module is used to calculate the corresponding compression parameters according to the input compressed image to be compressed and output to the JPEG compression module.
[0007] The JPEG compression module is used to compress the compressed image to be compressed according to the input compression parameters, and output the compressed image in the fixed code rate JPEG format.
[0008] In the image compression parameter generation module, a support vector machine trained and learned is used to classify the input compressed image to be compressed to obtain the compression parameters corresponding to the compressed image to be compressed.
[0009] Further, the image compression parameter generation module comprises a gray level co-occurrence matrix feature generation module, a support vector machine module, a compression parameter category coding storage module and a coding comparison module.
[0010] The gray level co-occurrence matrix feature generation module is used to generate the gray level co-occurrence matrix features of the compressed image to be compressed.
[0011] The support vector machine module is configured to obtain the code of the image to be compressed according to the input gray level co-occurrence matrix feature.
[0012] The compression parameter category code storage module is configured to store the codes of all compression parameter categories of the JPEG format image satisfying the fixed code rate after the support vector machine module is trained and learned.
[0013] The code comparison module is configured to compare the code of the image to be compressed with the codes of all compression parameter categories of the JPEG format image satisfying the fixed code rate, and determine the compression parameter category of the image to be compressed.
[0014] Further, the support vector machine module includes M classifiers; and each compression parameter category of the JPEG format image satisfying the fixed code rate corresponds to an M-bit code after the support vector machine is trained and learned.
[0015] Further, the gray level co-occurrence matrix feature extraction process of the gray level co-occurrence matrix feature generation module includes:
[0016] The gray level co-occurrence matrix set of the image is obtained by respectively counting the gray level co-occurrence matrices of multiple distances from multiple directions at different angles of the image to be compressed.
[0017] Multiple features are respectively extracted for any gray level co-occurrence matrix in the gray level co-occurrence matrix set.
[0018] All the extracted features are spliced to obtain the gray level co-occurrence matrix feature of the image to be compressed.
[0019] Further, in the gray level co-occurrence matrix feature extraction process, the gray level co-occurrence matrix set of the image is obtained by respectively counting the gray level co-occurrence matrices of 10 distances from four directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees.
[0020] Further, four features of contrast, correlation, energy and homogeneity are respectively extracted for each gray level co-occurrence matrix.
[0021] Further, the code comparison module calculates the Euclidean distance between the code of the image to be compressed and the code of any compression parameter category of the JPEG format image satisfying the fixed code rate, and selects the compression parameter category with the minimum Euclidean distance as the category of the image to be compressed.
[0022] Further, the image compression parameter generation module further includes a compression parameter optimization module.
[0023] The compression parameter optimization module is configured to reduce the compression parameter in the selected compression parameter category with the minimum Euclidean distance by 1 to obtain an optimized compression parameter for image compression.
[0024] Further, the JPEG compression module acquires the to-be-compressed image from the image acquisition module and acquires the compression parameter from the image compression parameter generation module to perform image compression; specifically including:
[0025] The to-be-compressed image is divided into blocks to obtain a plurality of source images, and the to-be-compressed image and resource allocation are initialized;
[0026] Each source image is divided into 8x8 data blocks in the order from left to right and from top to bottom;
[0027] The plurality of data blocks of any source image are transformed into discrete cosine transform coefficients through discrete cosine transform according to the compression coefficient;
[0028] The discrete cosine transform coefficients are quantized, and the quantized discrete cosine transform coefficients are Huffman entropy coded, and the compression image data corresponding to each source image is output according to the Huffman entropy coding;
[0029] The compression image data corresponding to the plurality of source images are spliced to complete the JPEG image compression of the to-be-compressed image.
[0030] Further, the training and learning process of the support vector machine in the image compression parameter generation module includes:
[0031] 1) Extracting the gray level co-occurrence matrix features of each training image in the training set;
[0032] 2) Dividing the N classes of training images in the training set M times based on the multi-classification task to obtain M training image subsets;
[0033] 3) For any training image subset, the gray level co-occurrence matrix features of each training image in the training image subset are input into the support vector machine for training and learning to obtain M classifiers.
[0034] The present application can achieve the following beneficial effects:
[0035] The present application can guarantee the compression rate of the image compression algorithm compression result, take into account the image transmission efficiency and the quality of the compressed image, and can minimize the loss of image quality, meet the power consumption, volume, transmission bandwidth, processing resources and other conditions in the application. BRIEF DESCRIPTION OF DRAWINGS
[0036] The accompanying drawings are included to provide a further understanding of the embodiments, and are incorporated in and constitute a part of this specification, illustrate embodiments of the application, and together with the description serve to explain the principles of the application, and should not be considered limiting of the application's scope, in which like reference numerals designate like components in the figures.
[0037] Figure 1 The composition block diagram of the device for performing fixed-rate JPEG format image compression in the embodiments of the present application is shown in the figure;
[0038] Figure 2 A block diagram of a component of an image compression parameter generation module in an embodiment of the application. DETAILED DESCRIPTION
[0039] The preferred embodiments of the present application will be described herein below with reference to the accompanying drawings, in which:
[0040] It should be noted that the embodiments and features of the embodiments in the present application can be combined if there is no conflict. The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. The description of the at least one exemplary embodiment is actually only illustrative, but not intended to limit the present application and its application or use in any way. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of the present application.
[0041] It should be noted that the terms used herein are only intended to describe specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form, unless the context clearly indicates otherwise, and it should be understood that, when the terms "comprise" and / or "include" are used in the specification, there is a presence of the features, steps, operations, devices, components and / or combinations thereof.
[0042] Unless specifically stated otherwise, the relative arrangements of the components and steps illustrated in these embodiments and the numerical expressions and values set forth herein are not limiting. It should be understood that the various parts of the drawings are not necessarily drawn to scale, and that, for the purpose of convenience and clarity, not all components can be shown in a given figure. Techniques, methods, and devices known to those of ordinary skill in the art can not be discussed in detail, but rather can be assumed to be known by those of ordinary skill in the art. In the examples shown and discussed herein, any specific values should be interpreted as merely illustrative, and not as a limitation on the scope of the exemplary embodiments. Thus, other examples of the exemplary embodiments can have different values. It should be noted that like reference numerals and letters refer to like items in the following drawings, and thus, once an item is defined in one drawing, it need not be discussed further in subsequent drawings.
[0043] An embodiment of the present application discloses an apparatus for carrying out fixed code rate JPEG format image compression, the fixed code rate requirement of image transmission is selected after comprehensive consideration of the signal bandwidth of compressed image transmission and the quality of transmitted pictures, and in the embodiment, PEG format image compression is carried out with ten times of image compression as an example.
[0044] As shown in the figure, the apparatus for carrying out fixed code rate JPEG format image compression comprises an image acquisition module, an image compression parameter generation module and a JPEG compression module. Figure 1
[0045] The image acquisition module is used for loading a to-be-compressed image for fixed code rate JPEG format compression.
[0046] The image compression parameter generation module is used for calculating corresponding compression parameters according to the inputted to-be-compressed image and outputting the compression parameters to the JPEG compression module.
[0047] The JPEG compression module is used for carrying out JPEG compression on the to-be-compressed image according to the inputted compression parameters and outputting a compressed image in JPEG format with fixed code rate.
[0048] In the image compression parameter generation module, a support vector machine trained and learned is adopted to classify the inputted to-be-compressed image and obtain the compression parameters corresponding to the to-be-compressed image.
[0049] Specifically, as shown in the figure, the image compression parameter generation module comprises a gray level co-occurrence matrix feature generation module, a support vector machine module, a compression parameter category code storage module and a code comparison module. Figure 2
[0050] The gray level co-occurrence matrix feature generation module is used for generating the gray level co-occurrence matrix feature of the to-be-compressed image.
[0051] The support vector machine module is used for obtaining the code of the to-be-compressed image according to the inputted gray level co-occurrence matrix feature.
[0052] The compression parameter category code storage module is used for storing all codes of the compression parameter category satisfying the fixed code rate JPEG format image compression obtained after the support vector machine module is trained and learned.
[0053] The code comparison module is used for comparing the code of the to-be-compressed image with all codes of the compression parameter category satisfying the fixed code rate JPEG format image compression, and determining the compression parameter category of the to-be-compressed image.
[0054] Specifically, the gray level co-occurrence matrix feature extraction process of the gray level co-occurrence matrix feature generation module comprises,
[0055] 1) respectively from the different angles of the image to be compressed in multiple directions, statistics of multiple distance gray level co-occurrence matrix to obtain the gray level co-occurrence matrix set of the image;
[0056] 2) for any gray level co-occurrence matrix in the gray level co-occurrence matrix set, respectively extracting multiple features;
[0057] 3) all the extracted features are spliced to obtain the gray level co-occurrence matrix feature of the image to be compressed.
[0058] Preferably, in the gray level co-occurrence matrix feature extraction process, respectively from 0 degrees, 45 degrees, 90 degrees and 135 degrees, 10 distance gray level co-occurrence matrices are respectively counted in each direction to obtain the gray level co-occurrence matrix set of the image.
[0059] Preferably, each gray level co-occurrence matrix respectively extracts four features of contrast, correlation, energy and homogeneity.
[0060] A preferred gray level co-occurrence matrix feature extraction process in this embodiment includes:
[0061] Respectively from 0 degrees, 45 degrees, 90 degrees and 135 degrees, 10 distance gray level co-occurrence matrices are respectively counted in each direction to obtain the gray level co-occurrence matrix set of the image, and the gray level co-occurrence matrix set includes 40 gray level co-occurrence matrices;
[0062] Each gray level co-occurrence matrix respectively extracts four features of contrast, correlation, energy and homogeneity.
[0063] The 160 features are spliced to obtain the gray level co-occurrence matrix feature of the GLCM-160 feature.
[0064] Specifically, the support vector machine module includes M classifiers; after support vector machine training and learning, the M classifiers correspond to an M-bit code for each JPEG format image compression parameter category that meets the code rate.
[0065] Preferably, in the support vector machine module, error correction output code (ECOC) is used to realize the multi-classification task of the input image to be compressed, and the inference stage is a decoding process. The feature of the image to be compressed is submitted to the M classifiers to obtain the code of the image to be compressed.
[0066] Specifically, the encoding comparison module calculates the Euclidean distance between the encoding of the to-be-compressed image and the encoding of the compression parameter category of the compression parameter category encoding storage module, and selects the compression parameter category with the minimum Euclidean distance as the category of the to-be-compressed image.
[0067] Preferably, the image compression parameter generation module further comprises a compression parameter optimization module.
[0068] In the test process of the support vector machine (SVM) classification result that the to-be-compressed image can be compressed by ten times, the misclassification phenomenon between adjacent categories may occur due to the proximity of features between similar categories (such as actually 5 misclassified as 6 or 4), so as to ensure that the compressed image will not be larger than the channel bandwidth, and the image compression parameter generation module is increased.
[0069] The image compression parameter generation module is configured to reduce 1 from the compression parameter in the selected compression parameter category with the minimum Euclidean distance to obtain an optimized compression parameter for image compression.
[0070] Specifically, the JPEG compression module obtains the to-be-compressed image from the image acquisition module, and obtains the compression parameter from the image compression parameter generation module for image compression.
[0071] The compression process in the JPEG compression module specifically comprises:
[0072] 1) The to-be-compressed image is divided into a plurality of source images;
[0073] 2) The plurality of source images are sequentially divided, and any source image can be divided into a plurality of data blocks;
[0074] 3) The plurality of data blocks of any source image are transformed into discrete cosine transform coefficients according to the compression coefficient through discrete cosine transform;
[0075] 4) The discrete cosine transform coefficients are quantized, and the quantized discrete cosine transform coefficients are Huffman entropy encoded, and the compression image data corresponding to each source image is output according to the Huffman entropy encoding;
[0076] 5) The compression image data corresponding to the plurality of source images is spliced to complete the JPEG image compression of the to-be-compressed image.
[0077] Preferably, the compression process is:
[0078] 1) The to-be-compressed image is divided into a plurality of source images, and the to-be-compressed image and resource allocation are initialized;
[0079] 2) Each source image is divided into 8x8 data blocks in the order from left to right and from top to bottom;
[0080] 3) Transforming the plurality of data blocks of any source image into discrete cosine transform coefficients by discrete cosine transform according to the compression coefficient;
[0081] 4) Quantizing the discrete cosine transform coefficients and Huffman entropy encoding the quantized discrete cosine transform coefficients, and outputting the compressed image data corresponding to each source image according to the Huffman entropy encoding;
[0082] 5) Splicing the compressed image data corresponding to the plurality of source images to complete the JPEG image compression of the to-be-compressed image.
[0083] The compressed image output by the JPEG compression module is a compressed image with a basic image compression rate of 10 times, which meets the channel transmission requirement while maximizing the quality of the compressed image.
[0084] Since the support vector machine in the image compression parameter generation module is well trained and learned, the training and learning process of the support vector machine in the embodiment includes:
[0085] The training set construction step: in the constructed training set, a plurality of training images and compression parameter categories corresponding to the training images are included; the compression parameter categories meet the requirement that the compression multiple meets the fixed code rate requirement when the corresponding training image is compressed and simulated;
[0086] The machine learning step: extracting the gray level co-occurrence matrix features of each training image in the training set, training the support vector machine to obtain M classifiers; after training, each compression parameter category corresponds to an M-bit code;
[0087] In the training set construction step, all compression parameters 1 to 100 are sequentially subjected to JPEG image compression simulation for each training image, and the signal-to-noise bandwidth of transmission and the picture quality of transmission are comprehensively considered, and the N compression parameter categories corresponding to the image compression of ten times are obtained from the image after compression simulation as the category of the training image. The training image before image compression and the compression parameter category corresponding to the training image constitute the training set.
[0088] In the machine learning step, it includes:
[0089] 1) Extracting the gray level co-occurrence matrix features of each training image in the training set;
[0090] 2) Dividing the N categories of training images in the training set M times based on the multi-classification task to obtain M training image subsets;
[0091] 3) For any training image subset, input the gray level co-occurrence matrix features of each training image in the training image subset into the support vector machine for training and learning to obtain M classifiers.
[0092] Specifically, the gray level co-occurrence matrix feature extraction process for each image in the training set includes,
[0093] Respectively from multiple directions of different angles of the training image, a gray level co-occurrence matrix set of the image is obtained by counting gray level co-occurrence matrices of multiple distances.
[0094] A plurality of features are extracted for any gray level co-occurrence matrix in the gray level co-occurrence matrix set.
[0095] All the extracted features are tensor spliced to obtain the gray level co-occurrence matrix feature of the image.
[0096] Preferably, the gray level co-occurrence matrix feature extraction process includes:
[0097] Respectively from four directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees, a gray level co-occurrence matrix set of the image is obtained by counting gray level co-occurrence matrices of 10 distances (d from 1-10) in each direction, and the gray level co-occurrence matrix set includes 40 gray level co-occurrence matrices.
[0098] Four features of contrast, correlation, energy and homogeneity are extracted for each gray level co-occurrence matrix.
[0099] The 160 features are tensor spliced to obtain the gray level co-occurrence matrix feature of the GLCM-160 feature.
[0100] Specifically, the error-correcting output code (ECOC) is used to implement the multi-classification task, and the training process is the encoding process; on the basis of the training set, the gray level co-occurrence matrix features of each training image in the training set are extracted; the N classes of training images in the training set are divided M times by using the error-correcting output code to obtain M training image subsets.
[0101] Specifically, in the support vector machine training and learning process, for any training image subset, the training images in the training image subset are divided, wherein a part of the training images are used as positive examples, and another part of the training images are used as negative examples; the gray level co-occurrence matrix features of the training images as positive examples and the gray level co-occurrence matrix features of the training images as negative examples are input into the support vector machine for training and learning to obtain the classifier corresponding to the training image subset.
[0102] After the support vector machine training and learning of all the training image subsets are completed, M classifiers are obtained, so that the trained support vector machine corresponds to one M-bit code for each compression parameter class.
[0103] The M-bit code obtained is stored in the compression parameter category code storage module.
[0104] To sum up, the device for carrying out fixed code rate JPEG format image compression provided by the embodiment of the application realizes fixed code rate JPEG format image compression of the image to be compressed, and compared with the prior art, the device can guarantee the compression rate of the compression result of the image compression algorithm, takes into account the image transmission efficiency and the quality of the compressed image, can minimize the loss of image quality, and meets the power consumption, volume, transmission bandwidth, processing resource and other condition limitations in the missile-borne application.
[0105] The above merely describes a preferred embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.
Claims
1. An apparatus for performing rate-distortion JPEG image compression, characterized by: The image acquisition module, the image compression parameter generation module and the JPEG compression module are included. The image acquisition module is configured to load an image to be compressed for fixed-rate JPEG format compression. The image compression parameter generation module is configured to calculate corresponding compression parameters according to the input image to be compressed and output the compression parameters to the JPEG compression module. The JPEG compression module is configured to compress the image to be compressed according to the input compression parameters and output a compressed image in the fixed-rate JPEG format. In the image compression parameter generation module, a support vector machine trained and learned is used to classify the input image to be compressed to obtain the compression parameters corresponding to the image to be compressed. The image compression parameter generation module includes a gray level co-occurrence matrix feature generation module, a support vector machine module, a compression parameter category encoding storage module, an encoding comparison module and a compression parameter optimization module. The gray level co-occurrence matrix feature generation module is configured to generate the gray level co-occurrence matrix features of the image to be compressed. The support vector machine module is configured to obtain the encoding of the image to be compressed according to the input gray level co-occurrence matrix features. The support vector machine module includes M classifiers. After the support vector machine is trained and learned, each fixed-rate JPEG format image compression parameter category that meets the fixed rate corresponds to an M-bit encoding. In the support vector machine module, an error correction output code is used to implement the multi-classification task of the input image to be compressed. In the inference stage, the decoding process is used to submit the features of the image to be compressed to the M classifiers to obtain the encoding of the image to be compressed. The compression parameter category encoding storage module is configured to store the encodings of all fixed-rate JPEG format image compression parameter categories obtained after the support vector machine module is trained and learned. The encoding comparison module is configured to calculate the Euclidean distance between the encoding of the image to be compressed and the encoding of any fixed-rate JPEG format image compression parameter category that meets the fixed rate, and select the compression parameter category with the smallest Euclidean distance as the category of the image to be compressed. The compression parameter optimization module is configured to reduce 1 from the compression parameters in the selected compression parameter category with the smallest Euclidean distance to obtain the optimized compression parameters for image compression.
2. The apparatus for performing rate-distortion JPEG image compression according to claim 1, wherein, The gray level co-occurrence matrix feature extraction process of the gray level co-occurrence matrix feature generation module includes: Statistically obtaining a set of gray level co-occurrence matrices of the image from multiple directions at different angles of the image to be compressed. Extracting multiple features from any gray level co-occurrence matrix in the set of gray level co-occurrence matrices. Tensor splicing all the extracted features to obtain the gray level co-occurrence matrix features of the image to be compressed.
3. The device for fixed-rate JPEG format image compression according to claim 2, wherein In the gray level co-occurrence matrix feature extraction process, the set of gray level co-occurrence matrices of the image is obtained from 10 distances in four directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees, respectively.
4. The device for fixed-rate JPEG format image compression according to claim 3, wherein Each gray level co-occurrence matrix extracts four features of contrast, correlation, energy and homogeneity, respectively.
5. The apparatus for performing rate-distortion JPEG image compression according to claim 1, wherein, the JPEG compression module acquires the image to be compressed from the image acquisition module and acquires the compression parameters from the image compression parameter generation module to perform image compression; specifically including: performing block division on the image to be compressed to obtain a plurality of source images, and initializing the image to be compressed and resource allocation; dividing each source image into 8x8 data blocks in the order from left to right and from top to bottom; transforming a plurality of data blocks of any source image into discrete cosine transform coefficients through discrete cosine transform according to the compression coefficients; quantizing the discrete cosine transform coefficients and performing Huffman entropy coding on the quantized discrete cosine transform coefficients, and outputting the compressed image data corresponding to each source image according to the Huffman entropy coding; splicing the compressed image data corresponding to the plurality of source images to complete the JPEG image compression of the image to be compressed.
6. The apparatus for performing rate-distortion optimized JPEG compression of claim 1-5, wherein, the training and learning process of the support vector machine in the image compression parameter generation module includes: 1) extracting the gray level co-occurrence matrix features of each training image in the training set; 2) dividing the N classes of training images in the training set M times based on the multi-classification task to obtain M training image subsets; 3) for any training image subset, input the gray level co-occurrence matrix features of each training image in the training image subset into the support vector machine for training and learning to obtain M classifiers.
Citation Information
Patent Citations
Image feature information intra-frame rapid division method and system, and storage medium
CN111259826A