局部傅里叶变换（LFT）是一种重要的纹理特征提取方法.该文首先分析了LFT系数各阶矩的纹理鉴别性能,然后采用特征空间中由不同纹理形成的聚类之间的Fisher判据作为评价指标,研究了纹理特征的鉴别性能,并通过纹理分割实验来验证相关结论,从而使结论具有更高的可靠性.分析结果表明,纹理图像的LFT系数一般不服从正态分布,其偶数阶矩具有较好的纹理鉴别性能,而奇数阶矩的纹理鉴别性能则较差,由于以LFT系数的2、4、6阶矩作为纹理特征,其纹理鉴别性能和分割结果优于Yu Hui和Haralick提出的纹理特征,因此,建议选用LFT系数的各偶数阶矩作为纹理特征.

Local Fourier transform(LFT) is an important texture feature extracting method.In this paper,the texture discrimination performance of LFT coefficients' moments is studied.The Fisher criterion is used to evaluate the discrimination performance of different texture features.Texture image segmentation experiments enable the conclusions more reliable.It is revealed that the LFT coefficients of texture images usually do not obey normal distribution,their even-order moments exhibit preferable texture discrimination performance,while their odd-order moments exhibit baddish.If LFT coefficients' 2nd,4th and 6th order moments are used as texture features,better texture discrimination performance and segmentation results can be achieved than those of Yu Hui and Haralick.So,it is suggested that the even-order moments of LFT coefficients are selected as texture features.