Development of Double Stage Filter (DSF) for Stereo Matching Algorithms and 3D Vision Applications

Abstract

A part of the stereo matching algorithms development is mainly focused on overcoming unwanted aspects such as noises, unwanted regions and occlusions. In this paper, a new technique which is called Double Stage Filter (DSF) is introduced. This technique is a hybrid algorithm which consists of dynamic programming and block matching. The main feature of DSF is mainly its function at the post-processing stage that is to remove the noises and horizontal stripes, obtained from the raw disparity depth map of dynamic programming. In order to remove the unwanted aspects, a two-stage filtering process is applied. In this DSF algorithm, segmentation process is also required to segment the optimized raw disparity depth map into several parts according to the pixel colours. The first filter block is applied to remove the noises of the segmented parts before merging. Meanwhile, the second filter is used to remove the unwanted region of the outliers on segmented parts after merging processes. The new disparity depth map of DSF is evaluated in Middlebury Stereo Vision page with a few evaluation functions, such as similarity structural (SSIM), peak to signal noise ratio (PSNR) and mean square errors (MSE). At the end of this paper, the performance of DSF is compared with other techniques.