Sparse matrix and convolutional neural network-based picture compression method

Abstract

The invention discloses a sparse matrix and convolutional neural network-based picture compression method. The method comprises the following steps of converting an image into two-dimensional matrix data; assessing, optimizing and constructing a sparse matrix image by utilizing a convolutional neural network; optimizing a compression scheme through the convolutional neural network; performing compression processing on the sparse matrix image by using the selected optimal compression scheme; and finally obtaining a high-compression-ratio and low-distortion compressed image. According to the method, the problem of difficult use and identification of the image due to low compression ratio and excessively reduced quality of the compressed image in an existing compression technology is solved; and the optimization can be carried out according to picture contents, and the proper image construction and compression method can be self-selected, so that the picture feature extraction and compression method has self-optimization and learning capabilities.

Description

technical field [0001] The invention relates to the technical field of picture processing, in particular to a picture compression method based on a sparse matrix and a neural network. Background technique [0002] Image compression technology is an important technical field in image processing technology. Image compression can effectively reduce file size, save storage space, and reduce network transmission pressure. [0003] At present, the common compression technologies include JPG, PNG, RAR, etc. JPG, PNG and other algorithms are specially designed for image format file compression, which can compress the original bitmap to 10%-70% of the size, and have been widely used in the computer industry. , the operating system and application software usually use these algorithms as the general standards supported by default, but the disadvantage is that these two algorithms will have different degrees of quality loss according to different compression ratios. RAR and other tech...

AI Technical Summary

Problems solved by technology

[0004] At present, the compression ratio of most compression algorithms in the industry is not high enough. Usually, the original bitmap can be compressed to 10%-70% in size, and a few methods can be compressed to 1%-5%. The content is limited or the quality of the image has dropped a lot

The image compression algorithm used in some specific industries has an efficient compression ratio, but there is a problem that the quality drops too much after compression. After the image is compressed, the quality drops too much, making it difficult to use and identify.

[0005] The image compression algorithms used in some specific industries have high compression ratios and good quality, but they cannot adapt the algorithm according to the image content.

It can only be optimized for special pictures of a particular industry and the rules of pictures in this industry. This type of algorithm limits the content and application scenarios of pictures, and cannot perform efficient and high-quality compression on pictures of any content.

For example, when taking pictures of the night sky in the field of astronomy, there are only a few bright stars in the picture, and the rest are dark in color. According to the characteristics of this type of picture, a specific algorithm can be compiled for efficient compression, but this algorithm cannot be extended to pictures with any content.

[0006] To sum up, none of the existing image compression systems and methods have a system and method that has an extremely efficient compression ratio, less quality degradation after compression, and can intelligently improve the algorithm according to different images

Images