关键词: 傅里叶描述子, 椭圆系数, 多尺度, 形状表示

Abstract: Shape representation is one of the most important research contents in the field of pattern recognition and computer vision. Considering that the traditional shape representation algorithm cannot describe the whole characteristics and the detail information well at the same time and the versatility is also not desired, a new elliptic Fourier descriptor based on the Gaussian multiscale analysis is proposed in this paper. This algorithm makes convolution between Gauss function and complex coordinate function of the target object. Through the choice of parameters of Gaussian curve, the boundary information can be presented into different scale spaces. And then it can get a shape characteristic vector through elliptic Fourier transform. When this method is used to describe the shapes of a same kind, the correlation coefficients between the characteristic vectors are very high. On the contrary, the coefficients are very low when describing the shapes of different kinds. The experimental results show that this method has good translation, rotation and scale invariance, strong shape discrimination ability and more accurate results in the shape classification and recognition experiment.

Key words: Fourier descriptor, elliptic coefficient, multiscale, shape representation