Abstract

In this paper, we propose a method for image compression in which currently developed vision tech-nologies are incorporated with traditional trans-form-based coding methods. Based on newly developed vision technologies, we present two coding-oriented in-painting methods: structure-aware inpainting and ob-ject-based inpainting. Examples show the effectiveness of these inpainting methods on information recovery employing certain kinds of assistant information. More-over, a primary image compression system has been built to test the proposed compression approach in which some regions are intentionally removed at the encoder only to be restored naturally at the decoder. Up to a 50% bit savings can be achieved at similar visual quality lev-els when compared with JPEG.