| Summary: | 碩士 === 國立高雄第一科技大學 === 電腦與通訊工程研究所 === 101 === This work proposes an algorithm that can effectively spot and recognize Arabic numbers from a sequence of fingertip trajectory images. The use of chain code probability (CCP) to represent the fingertip trajectory generally suffers from scaling and code accumulation problems. To handle these two problems, this work proposes a fingertip representation called local CCP (LCCP) which imposes local property to traditional CCP. As for spotting and recognizing fingertip trajectory, the windows with different lengths are applied to scan over the entire video sequence. The segments within the window have distance using template matching are considered as effective Arabic numbers. However, the same segment may correspond to several different Arabic numbs which is referred to as overlap problem. Therefore, the inclusion relation among Arabic numbers is used to resolve the overlap problem. In experiments, we validated our proposed algorithm by using 40 videos. The results show that the proposed algorithm using LCCP and inclusion relation outperforms the other ones in spotting and recognition accuracy.
|
|---|
