Shape Representation

• Main Authors: Shou-Ping Shih, 石受平
• Other Authors: Leu-Shing Lan
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/33000685846219895856

碩士 === 國立雲林科技大學 === 電子工程與資訊工程技術研究所 === 87 === Shape representation is essential to image description and shape recognition. It encompasses wide applications, such as video and image compression, pattern recognition, computer graphics, multimedia communications, digital library, image database, etc. In this thesis, we present our research results related to this subject. Two new shape representation methods are proposed. The first, called peak-error-constrained B-spline representation, reduces the large peak errors that sometimes occur in the traditional B-spline technique. We have developed two optimization algorithms to solve this peak-error-constrained minimization problem. Both of them converge fast（typically below 5 iterations）. The second method, cosine descriptor, is a close relative to the traditional Fourier descriptor. However, it is better than the latter because it can be easily extended to N-dimensional shape representation and does not require complex multiplication. In another part of our research, we performed error analysis of the traditional B-spline method. The result is summarized in the form of an empirical formula which relates the quantity of error to the compression ratio. This formula is useful in the determination of the number of control points in the B-spline method.