Classification of Face Contours Using Genetic Algorithm

• Main Authors: Jia-Rong Zhuang, 莊家榮
• Other Authors: Jiann-Der Lee
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/98046366786931245436

碩士 === 長庚大學 === 電機工程研究所 === 88 === A major problem in designing highly specialized equipment, such as oxygen masks, is that the effectiveness of the equipment depends on its appropriateness for the size and shape of the body part for which it is designed. In general, among the individuals who are likely to be using this equipment, there is considerable heterogeneity in size and shape of the body parts. One solution is to use available data to form homogeneous clusters of the population and then make separate designs for each cluster, commonly referred to as sizing. However, for certain complex designs, this sizing information, while useful, is insufficient. In this thesis, we proposed two genetic-based methods to solve the problems of classifying biological shapes, especially face contours, and to find out standard models for designing specialized equipments. A face contour is first transformed to sequent signals by an orientation function The continuous wavelet transform is adopted to preserve the orientation property both in time domain and frequency domain in different scales. The characteristic of multiple resolutions is helpful to extract the dominant control points of a face contour and to construct a feature set. These feature sets are regarded as initial states for classification. With the obtained initial states, two algorithms, the “Weighted Genetic Algorithm” and “Integrated Genetic Algorithm”, which are improved from the traditional genetic algorithm, are proposed for classification. The algorithms completed by finding out the appropriate centroids of classes. The classification correctness is higher than 94.44%, and fitted in well with the human perspective.