Perceptual-oriented Learned Image Compression with Dynamic Kernel

• DOI: 10.48550/arxiv.2401.13967
• Publication Date: 2024-01-01
• Authors: Nianxiang Fu; Junxi Zhang; Huairui Wang; Zhenzhong Chen
• Keywords: Kernel (algebra)

In this paper, we extend our prior research named DKIC and propose the perceptual-oriented learned image compression method, PO-DKIC. Specifically, adopts a dynamic kernel-based residual block group to enhance transform coding an asymmetric space-channel context entropy model facilitate estimation of gaussian parameters. Based on DKIC, PO-DKIC introduces PatchGAN LPIPS loss visual quality. Furthermore, maximize overall perceptual quality under rate constraint, formulate challenge into constrained programming problem use Linear Integer Programming method for resolution. The experiments demonstrate that proposed can generate realistic images with richer textures finer details when compared state-of-the-art techniques.