Inpainting intergrate with decomposition for image compression

Abstract

In this paper, we propose an image compression framework that integrates image decomposition technique with image inpainting to exploit visual redundancy in color images. An original image is first decomposed into geometry and texture using nonlinear filter decomposition model. Then, the geometry and texture image are compressed with edge-based inpainting and the standard JPEG respectively. At the decoder side, the decoded texture image adds to geometry image which is decoded using structure propagation and texture synthesis. In this scheme, we study different decomposition models and choose the best decomposition model to get pure texture image. Moreover, robustness of texture image is fully taken advantage. Based on that, more regions can be dropped at the encoder side without sacrificing the image quality. Experimental results show that our method significantly outperforms standard JPEG and image compression with edge-based inpainting.