An Image Inpainting Method for Stereoscopic Images Based on Hole Classification

Abstract

This paper proposes a image in painting method for stereoscopic images based on hole-classification. The stereoscopic image pairs and their depth map images are generated by left and right shifting the original image according to its corresponding depth information. The generated left and right image may have disocclued areas or holes in the image after shifting operation. In this paper, disocclued areas or holes are classified into three type, gradient, clear-cut and layer and three different filling routes and filling methods are adopted according to these three types of holes. From the experimental results, it is proved that holes maybe mistakenly filled have been in painted with better corrections, in the meantime, in painted image becomes more natural and closer to the reality when it is viewed.