Image Compression Using Deep Convolutional Adversarial Networks

Image compression is a kind of compression of data, which is used to images for minimizing its cost in terms of storage and transmission. Neural networks are supposed to be good at this task. One of the major problem in image compression is long-range dependencies between image patches. There are mainly two approaches to solve this problem; one is to develop a better residual patch-based encoder, and the second one is to create an entropy coder capable of collecting long-term dependencies inside the picture between patches. We address both the problems in this paper and fuse the two possible solutions to improve compression levels for a given material. Results of the simulation reveal that the new algorithm works much better in all parameters than in the current model.