Multiscale based shape representation and recognition

• Main Authors: Guan-Shu Tseng, 曾冠樹
• Other Authors: Chin-Hsing Chen，Yung-Nien Sun
• Format: Others
• Language: en_US
• Published: 1994
• Online Access: http://ndltd.ncl.edu.tw/handle/07067139102213738822

碩士 === 國立成功大學 === 電機工程研究所 === 82 === In this paper, a new shape representation and recognition method is proposed to identify single objects or overlapping objects from model objects. By using the wavelet transform, the boundary of an input object is decomposed into three signals, from which a new feature points selection and feature vectors assignmen scheme is used to extract magnitude and position change information. The shape recognition task is achieved through feeding these feature vectors of the feature points belonging to the input object and the model into a modified Hopfield neural network. The modified Hopfield neural network is employed to perform global and parallel feature matching. The modification of the network aim at achieving unique and stable matching result. Experiments were conducted to illustrate the performance of the proposed method for single or overlapping objects recognition.