A hole-filling algorithm based on pixel labeling for DIBR

Abstract

Depth Image Based Rendering (DIBR) technology is one of effective methods to generate stereoscopic image pairs in 3D image warping, however, the holes would be produced when using this method. Hole-filling algorithms are essential for improving image quality of stereoscopic image pairs. In this paper, a new hole-filling algorithm based on pixel labeling is proposed. Firstly, holes in stereoscopic image pairs produced by DIBR are marked as 0, whereas marked as 1. Then traversing the image pairs only once to fill pixel values of each hole according to the situation of hole’s eight neighborhood pixels, the hole would be filled by the average of no-hole pixel values when the number of no-holes greater than threshold, otherwise the hole is filled by the cross diamond search algorithm from every direction to find the closest no-holes until the number of no-holes greater than threshold. The proposed method is evaluated by existing objective assessment methods, such as PSNR and SSIM. Experiment results show that the proposed hole-filling algorithm provides an improvement in both of subjective and objective assessment by compared with the conventional hole-filling algorithm under the same source images. The proposed algorithm is not only simple, but also can effectively eliminate the holes generated by using the DIBR method.