Edge Detection Algorithm for Depth Map Generation in Free View Video

High-quality depth map generation is a key component of free-viewpoint video (FVV), because the virtual view-point images used for FVV are synthesized from real multi-view images and their depth maps. However, synthesized images do not guarantee good quality if the depth maps are inaccurate. Many studies have been conducted to develop methods for generating accurate depth maps. A semi-automatic algorithm was proposed, but it requires considerable time-intensive user involvement. Fully automatic methods have also been proposed, but their quality is inferior to the semiautomatic method. In this paper, we propose an edge detection algorithm for defining a precise object boundary to produce a better depth map. Better maps are generated compared with those produced by the automatic mode in Depth Estimation Reference Software (DERS) program. The peak signal-to-noise ratio (PSNR) of images synthesized by View Synthesis Reference Software (VSRS) based on this proposed algorithm was improved by about 0.2 dB. Key-words: Free-viewpoint video (FVV), Depth Image Based Rendering (DIBR), depth map, semi-automatic mode, edge map