Fast Affine Template Matching using Coarse-to-Fine Optimal Search with Distributed Sampling Points

• Main Authors: Yao-BinYang, 楊矅賓
• Other Authors: Jenn-Jier Lien
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/28613149533692220674

碩士 === 國立成功大學 === 資訊工程學系 === 103 === In recent years, algorithms of image analyses have been important since rise of human-computer interaction and industrial automation. Applications in these fields are about interactions with actual objects through image information for a specific purpose. How to accurately and quickly analyze necessary information has become a primary objective. In order to obtain information of a specific pattern in an image, template matching becomes an important technology. This thesis presents a solution to a template matching problem using an optimal search. A proposed method can accurately and quickly find locations, scales, and orientations of the specific pattern without analyzing image features. When information about a specific pattern and an image is obtained, a transformation set is approximately retrieved from infinite transformations. Then, the transformation set with sums of absolute differences is evaluated to judge whether to continue the optimal search under restrictions. During the optimal search, relatively poor transformations are removed. Then, the optimal search with the rest of the transformations is done in a small area to find new transformations. After the new transformations is found, evaluations, judgements, and fine searches are performed until a convergence or the maximum number of searches is achieved. Finally, the best transformation is optimally computed.