Recognitions of Tonal Sound and License Plate Character Using Filter-driven Template Matching
In this paper, we propose a filter-driven template matching scheme which is capable of recognizing one-dimensional signals and two-dimensional patterns. Our method is the improvement of the conventional template matching, meanwhile, it can be considered as a restoration process. We estimate a filter and then apply it to reference signal/pattern, which makes the filtered reference signal/pattern as similar as possible to the query one. In the literature, two common used approaches were available for filter estimation, namely least squares estimator (LSE) and quadratic programming (QP). Afterward, computing the errors between the filtered signals/patterns and the query one, one of filtered signals/patterns with minimum error is the best resultant of recognition. The proposed method is robust against several operations, including scaling/amplifying, filtering, translation and noise addition. Experiment results demonstrate that the filter-driven template matching scheme is superior to the conventional template matching in recognition of one-dimensional tonal sounds and two-dimensional license plate characters.
Yi-Chong Zeng Jian-Jiun Ding Soo-Chang Pei
Academia Sinica, Taipei National Taiwan University, Taipei
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-7
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）