Texture segmentation is one of fundamental problems in image processing. An efficient and robust method is needed for various texture samples. Consequently

this paper presents a texton-based algorithm of texture image segmentation. At first

Harr wavelet is utilized as a transform tool so that texture subimages of different directions are produced. And then

we use double threshold technology to get edges and propose a new method to extract textons of four directions. Thereafter

textons are simplified to one dimension vector and reduced to twelve kinds. At last

Statistics and vector field are applied respectively in order to separate texture images from coarse to fine. The algorithm is simple and effective for a wide of textures. Through this way

texture images are not only separated

but also texture structure can be described

which is useful for high-level image processing tasks.