提出了一种基于非乘积型小波的纹理分析方法：首先利用非乘积型小波分解原始图象，然后对高频部分的子图分别进行像素能量计算、纹理滤波以及直方图分割，从而得到粗分类的结果图；再把所有子图合并，并进行适当的后处理；最后利用小波的多尺度特性，决定分析是否终止，从而得到纹理分析结果。本方法是一种无监督分类，精确度高且计算复杂度低。

In this paper, we provide a wavelet transform method for texture analysis. We first decompose the image by a nonseparable discrete wavelet. By calculating the pixel energy of high frequency bands, we can extract the texture feature. The number of texture class is determined by coarse segmenting based on histogram. The final result can be obtained, after fine segmentation and multi scale processing. Our approach is an unsupervised classification algorithm, which is simple but accurate.