Stereo Matching Enhancement Algorithm Based on Fine Image-Structure Analysis

• Main Authors: Yu-Ching Chen, 陳昱磬
• Other Authors: Chan, Din-Yuen
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/9wb6w6

碩士 === 國立嘉義大學 === 資訊工程學系研究所 === 107 === In this thesis, we propose a scheme for improving the quality of parallax map by using the improved aggregation cost, smoothness constraint and penalty constraints. Firstly, the penalty constraint, which calculates the multiple aggregation costs within the maximum parallax through the reference map and the target map, and recalculating the new multiple aggregation cost through the penalty constraint of the fine structure. Then, using the smoothing constraint of the second fine structure to calculate the parallax value of the eight adjacent edges around the parallax map, and the new multiple aggregation cost is recalculated. Finally, the final disparity value of the pixel is obtained. Next is our proposed adaptive parallax search range, which will calculate the appropriate parallax search range according to each pair of images and the most similar partial window from the reference and target images. We will calculate it twice. At first, it will calculate from left to right, and secondly, from right to left. For each time, there will be a preliminary parallax search range value, and then we'll take the maximum one and multiplies it by an error value. Finally, using the left-right consistency(LRC) check. The parallax maps of the left and right image are merged, and the consistent parallax values are reserved, and the inconsistent parallax value will be replaced by the left-right consistency(LRC) check algorithm.