Visual coherence metric for evaluation of color image restoration

Digital image inpainting technique could provide substantial support for future restoration of images. However, much less effort has been devoted to the development of image inpainting evaluation. In this paper, we proposed a novel objective metric for assessing the quality of color image inpainting which takes into account some constraints and characteristics related to the specific objectives of inpainting approaches. The used characteristics are the visual coherence of the recovered regions and the visual saliency describing the visual importance of the area. A series of psychophysical experiments have been conducted to evaluate the performance of the proposed image quality index. The results show that the proposed image quality metric is well adapted for image quality inpainting evaluation.

1. INTRODUCTION

The goal of digital image inpainting is to fill the lost content automatically according to current image information in a visually plausible way, i.e., the restored parts are not easily detectable by viewers and not annoying. Digital inpainting has found widespread use in many applications such as image editing (removing undesired objects, restoring the scratches), film reproduction (deleting the logos, subtitles, etc), or even creating artistic effects (reorganizing objects, smart resizing of images, blending images), etc. In the field of archaeology, it can be employed for digital restoration of damaged objects such as statues, buildings, walls, etc.