A new texture description method of an image is proposed in the paper. Firstly, a binary image which reflects the image’s low frequency is defined by comparing the image’s gray mean with the region’s. Then, the statistic features are extracted based on the co-occurence matrix of the binary image. Finally, the new texture feature and the traditional texture spectrum are used as an index. The experimental results show that the new method is efficient and it provides noticeable improvement to the performance of texture spectrum descriptors.