Auto-registration of Fourier Transform Magnitude Spectra and its Application on Texture Identification and Segmentation
傅氏变换的自配准性质及其在纹理识别和图象分割中的应用

Texture identification and segmentation of images are very important issues in image analysis. In this paper we discuss the mechanism of using Fourier transform on texture identification and then suggest a texture image segmentation algorithm based on Fourier transform. Even though the edges of texture images may have various orientations and their locations in the image may be random, for the magnitude of Fourier transform of the image, the contribution of all edges with the same orientation will be stacked up in the orientation being perpendicular to the edges. This special phenomenon is called as auto registration of the magnitude spectra. The auto registration also means a re distribution of contributions of all patterns of the texture, according to the orientations of edges and the frequency locations of patterns rather than spatial locations of them. In this paper we illustrate and theoretically analyse the auto registration property of the magnitude spectra and then propose a method to exploit this property on texture identification and image segmentation. Experimental results show several advantages of this method. It is demonstrated that Fourier transform based method is capable and promising on texture identification and segmentation and a deeper research on this subject is worthwhile doing.