Multimedia Image Data Compression And Transmission Design

With the rise and development of computing technology,there is a large number of multimedia information such as images and videos on the network,occupying a large amount of network bandwidth.Accessing the platform through the network often faces problems caused by the high-bandwidth delay network environment,which affects the network transmission efficiency.Aiming at the problem of image transmission in multimedia networks,this thesis proposes the design of convolutional autoencoder image compression algorithm and the design of image compression algorithm based on GAN under the framework of deep learning theory to achieve the compression processing of the image.At the same time,by designing an adaptive transmission scheme,the transmission protocol optimization of the transparent proxy is added,and the response time is reduced according to the design of the TCP proxy,and the test is carried out in the real high-speed and long-delay network environment to achieve efficient transmission of multimedia image data.The main arrangements are as follows:(1)Analyze the problems of image transmission in multimedia networks.By studying and comparing the advantages and disadvantages of various image compression algorithms and transmission methods,the design of the image transmission optimization scheme in this thesis is determined,and the improvement ideas are proposed.(2)According to the design scheme,aiming at the problem that the traditional autoencoder fails to perform sufficient bit allocation and low performance,a content-weighted convolutional autoencoder network model framework is proposed,which reasonably controls the entropy rate,solves the problem of entropy rate optimization,and achieves better results than the traditional image compression method.Aiming at the problem that the image reconstruction quality is not high and the blur distortion is easy at low code rate,a generative adversarial network framework based on multi-scale discriminator is proposed,which minimizes the distortion of each scale through training with different weights,and obtains higher quality reconstruction images under low code rate conditions.The use of WGAN-GP avoids the problems of mode collapse and gradient vanishing.Experiments show that the model obtains better local texture and semantic information under low bitrate conditions,and realizes more efficient processing of image compression.Aiming at the problem of slow image transmission efficiency of TCP transmission protocol,an optimization design to increase transparent proxy is proposed,the packet retransmission process is optimized,the transmission efficiency is improved,and the transmission index analysis and optimization prove the effectiveness of the design.(3)According to the system requirements,the image compression design algorithm in this thesis is compared with other methods,which shows that the autoencoder design under convolutional network is weighted by content by importance graph,and image compression is realized on the basis of smaller bit rate。The GAN network of multi-scale discriminator realizes image reconstruction at low code rate by sampling data of different scales and continuously adversarily training through generators and discriminators.Finally,an experimental environment for image transmission testing is built,and the image compression method and TCP transmission optimization are experimentally verified.The results show that for image transmission,the compression method in this thesis not only has low distortion,but also has good effect at low bit rate,and the compression rate is further improved to make the transmission efficiency better.At the same time,the transmission system with the addition of transparent proxy reduces the garbage rate,the response time is also faster,and the transmission performance is better.