A Reference Resource Based End-to-End Image Compression Scheme

Abstract

Deep learning and convolutional neural networks have achieved a great success in computer vision and image processing, especially in low-level vision problems such as image compression. Recently, some end-to-end image compression methods have been proposed leading to a new direction of image compression. In this paper, we propose an end-to-end reference resource based image compression scheme to exploit the strong correlations with external similar images. In the proposed scheme, the side information is generated from highly correlated images in the reference resource. The features of side information can conceptually guide the compression process and assist the reconstruction process. The important map is employed to guide the allocation of local bit rate of the residual features. The proposed compression scheme is formulated as a rate distortion optimization problem in an end-to-end manner which is solved by ADAM algorithm. Experimental results prove that the proposed compression framework greatly outperforms several image compression frameworks.