Research On Image Compression Algorithm Based On Deep Learning

With the development of 5G,the massive amount of information and data has brought about the tension of storage devices and the stagnation of network resources.Therefore,seeking a reasonable and effective image compression method to improve the utilization of network resources will become one of the hot spots in the field of image research in the future.Deep learning currently has relatively mature applications in the image field.Using deep learning methods to explore more scientific and efficient image compression methods is bound to be one of the effective ways to solve the storage of massive image data.This paper studies image compression methods based on deep learning technology.Through the study and analysis of deep learning technology and its application status in the field of image compression,an image compression method based on multi-stage adversarial network is proposed.The compression accuracy and the resolution of the reconstructed image were optimized,and the image processing software based on the multi-stage anti-image compression method was designed.Realize the application of image compression method.The specific work is as follows:(1)Research and analysis of image compression algorithm based on deep learning,according to CNN-based image compression method,RNN-based image compression method and GAN-based image compression method for research.The research and analysis of representative examples of different methods are carried out,the commonly used evaluation indicators are summarized,and their future development trends are discussed and analyzed,which lays the foundation for follow-up research.(2)This paper presents an image compression method based on a multi-stage adversarial structure,which further reduces the amount of image compression data and realizes the optimization of the definition of the reconstructed image.A multi-level confrontation structure is constructed through an image pyramid,a generative confrontation network is introduced into each level of the pyramid structure,and an end-to-end image compression framework based on a three-layer structure is designed.The experimental results show that under the conditions of a given compression ratio,the multi-stage adversarial structure is significantly better than the classic method in reconstructing image clarity,and it has achieved better results in the comparison of PSNR and SSIM values.(3)This article uses Pycharm combined with Anaconda and PyQt5 to develop an image batch processing software based on image compression.The software consists of an image compression module,an image reconstruction module,an image quality processing module,an image watermarking module,and an image renaming module.Among them,the image compression module and image reconstruction module are the network models trained by the multi-stage confrontation image compression method in this article;the image quality processing module can handle image definition;the image watermarking module and image renaming module are used as common functions.The entire software interface is clear and clear,and the process operation is simple and clear,which can meet the needs of novice users for image processing.