Multi-view Stereoscopic Image Generation Based on DIBR with Stereo Image Pair

Abstract

Traditional stereo display technique is wearing glasses, but we make inconvenient for viewer. So autostereoscopic display be born. Besides it overcome the uncomfortable of wearing glasses, the effect of stereo image and similarity are better than traditional stereo displayer. Because it have high cost and high technical barriers to make Multi-view image, using software simulation making Multi-view image is easier and more economic. Depth-image-based Rendering (DIBR) is major Multi-view image algorithms. It is using single image and depth map to shift pixel and then computing other view image. However, it easily make visually disoccluded regions in shifting process. By above concept, this paper propose an improved padding method for DIBR. Our research using side by side-image to define the boundary of foreground and background, and find disoccluded regions. After producing virtual view image, finding the correspondence position of left eye view or right eye view by using percentage concept to compute the disoccluded regions. Finally using having information to pad. According the experimental results apply to autostereoscopic display, it is more accurate to hole filling and get better effect of stereo display than traditional padding method.