关键词: Fourier变换, 点匹配, 轮廓分解, 局部描述

Abstract: This paper proposed a strategy for retrieving binary images based on contour decomposition and local description. Firstly, the contour of object was extracted from binary image and decomposed by special method, and then the set of reference points were acquired. For each reference point, the curve to which the point corresponds was gained. A set of vectors which connect the reference point to all the points in the curve were computed. After that, Fourier transform was applied to the set of vectors and Fourier coefficients were treated as the eigenvector of the reference point. As a result, the image could be represented by a set of feature vectors in feature space. Finally, the distance between two images could be calculated by the method of points matching and the retrieval of binary images could be implemented. Experiments show that this method has higher retrieval precision, compared with some classical methods.

Key words: Fourier transform, point matching, contour decomposition, local description