Research On Deep Learning-based For Cloud Storage Of JPEG Images

Vision is the main way for people to perceive and recognize the outside world,while images are the most intuitive way of expressing objective things and the main way for people to obtain information.After entering the digital society,millions of images are transmitted to various social networking sites such as Facebook,Weibo,and We Chat every day.The digital multimedia information is massive in data.The JPEG image format is currently the most widely used image storage format.Most of the images on the network are stored in the JPEG format,so the compression of JPEG images has a high practical application value.For example,a JPEG image file is uploaded to a network disk,and the JPEG image file can be losslessly compressed to reduce the storage space,and decompressed during download to restore the original JPEG image.Our goal is to provide higher lossless compression for JPEG files in a compatible manner,which can reduce the file size of JPEG images and restore them to the original JPEG images.In this paper,we propose a cloud storage system to reduce the storage cost of JPEG images.Specifically,the uploaded JPEG image is decompressed to obtain the image height,width,quality factor,and quantized DCT coefficients.Then the image height,width and quality factor parameters are encoded,and the neural network-based arithmetic encoder is used to encode the quantized DCT coefficients to obtain a new code stream file and store it on the cloud.When the user downloads the image,the code stream file is decoded to obtain the image height,width and quality factor parameters.And an arithmetic decoder based on neural network is used to decode the quantized DCT coefficients.Then recompress these parameters and the quantized DCT coefficients to get the original JPEG image and send it to the user.Our system can compress JPEG images with low and medium bit rates.We test the average reduction rate of 24 images in Kodak dataset compared with the benchmark JPEG compression standard in different bit rate scenarios.The simulation results show that our system reduces the average storage space by 60.27% at the highest and 24.41% at the lowest.