Methods Of Texture Image Segmentation Based On Fractional Fourier Transform

Texture image segmentation has extensive applications in remote sensing,medical image processing,image database of texture-based and content-based retrieval and many other areas.In recent years,texture image segmentation is more and more important and becomes an indispensable step in image processing area,it is the basis of image analysis in practical application. There are four commonly used methods on description of texture:statistical analysis,stuctural analysis and spectrum analysis.As the most of texture have apparent spectrum feature and directionality,and texture segmentation requires that the expression of feature have locality.Therefore,the transform tool adopted in method of texture segmentation is required for high definition both in time domain and spectrum domain. Usually,the traditional method of we analysing and processing signal and systems is fourier transform,it is a very important tool.But with the deep researches and wide applications,fourier transform exposes its limitations in studying certain questions.After transformation we get the whole spectrum of signal,but not the partial feature.Therefore,fourier transform can only process the definite stable signal,for time-varying unstable signal is powerless.In order to overcome this disvantage of fourier transform in texture segmentation,the paper uses the Wigner-Ville distribution as a major transformation tool. The paper uses the basic properties of fractional fourier transform, use the Wigner-Ville distribution of cross-terms suppression methods in fractional domain and estimation of optimal fractional domain to two-dimensional texture image signal, so as to overcome the shortcoming of the locality caused by traditional Fourier transform based texture segmentation method, and successfully avoide the disturb of multi-component signal cross-terms in two-dimensional texture image's Power spectrum.Sampling the texture image segmentation, calculating the power spectrum without cross-terms. The statistical characteristics of the power spectrum-variance as vector texture characterization is used in texture clustering.The experimental simulation results show that, the new method can better express the regional feature of texture image, comparing to the traditional method of texture image segmentation based on fourier transform,additional,using K-means clustering algorithm can obtain more precise segmention affect。...