ELITE—A Novel Approach to Translation, Rotation, and Scale Invariant Template Matching

• Main Authors: Wang, Yi-Hua, 王奕驊
• Other Authors: Wang, Jung-Hua
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/h49be7

碩士 === 國立臺灣海洋大學 === 電機工程學系 === 102 === This thesis aims to match patterns having translation, rotation, and scale differences in an input image. Through integration of advantages of area-based matching (ABM) and feature-based matching (FBM) approaches, a novel pattern matching algorithm called Equalized Length Identification to Evaluation (ELITE) is proposed. ELITE mainly consists of a circular sampling process, interpolations of the sampled data, and establishment of estimation space. Unlike other circular sampling methods [14,15,16], in ELITE the circumference of shorter sampling circles is extended to a length equal to the longest sampling circles by interpolations. A continuous profile can be obtained by stacking multiple extended data in parallel, and interestingly, patterns of different scales would share a similar profile form with a gradual change, this correlation property is used to create a scale-estimation feature set effective to achieve both the rotation invariance and scale invariance. Also noteworthy is that such equalization profile extracted from a target pattern is just like a unique identification or fingerprint for the target pattern. It is not affected by the sampling range, and more desirably, it is identifiable and estimable. Experimental results show that, the proposed ELITE not only is able to locate or identify multiple patterns under influences of rotation and scaling, but also enables to access the rotational orientation and the scale of the input pattern. ELITE also exhibits a certain degree of tolerance to affine distortion.