Improved Census Transform’s Disparity Map by Using Edge Information

• Main Authors: Shih-Cian Huang, 黃士謙
• Other Authors: Jiun-Jian Liaw
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/13151524828878918932

碩士 === 朝陽科技大學 === 資訊與通訊系 === 104 === In the Stereo Vision system, baseline and focal are fixed, so how to get the disparity correctly is critical for obtaining the object distance. The Census Transform (CT) is an algorithm which can be used to solve the image matching problem, and it is more resistant to light change and higher robust. However, CT's huge demand of calculation quantity and memory has made CT difficult to be applied in the Real Time system. This paper found in the experiment that CT could not provide good transforming results in the area with insufficient high-frequency information. Therefore, this thesis was focused on designing Haar Wavelet's Modified Census Transform (MCT), Adaptive Window Census Transform (AWCT) and Adaptive Window Sparse Census Transform(AWSCT)with the change of the window size. Among them, MCT utilized the Wavelet Transform's characteristic of multiple layered analysis to separate the high and the low bands to modify CT's disparity and improve the correctness in the process of MCT's running in the small converting window. As far as AWCT transformed window was concerned, the results of CT's transform were applied to classify the image into the simple image with edge. In addition, the CT transformed window was changed from 3×3 to 21×21. Consequently, AWCT selected the proper size of the transform window, and enabled CT's transform window to change in accordance with each order condition without unnecessary calculations, while CT's hardware demand was decreased on the other hand. Experiments confirmed, our method can reduce unnecessary calculations make CT more faster and flexible.