Research Of Guided Smoothing And Segmentation On Texture Images

Many images contain textures.Smoothing and segmentation play a very important role in feature extraction of texture images.Due to the high color contrast of the texture,traditional methods of smoothing and segmentation based on contrast cannot get desirable results on texture images.Thus,this paper proposes guided smoothing and segmentation methods on texture images,where the computed guidance image controls the process of smoothing and segmentation to obtain better results.The main research contents include:(1)Most of the existing methods of texture smoothing utilize the statistical features of pixels within rectangular patches to distinguish the structure from the texture,where the patch sizes of rectangular are fixed.It is difficult to select a best patch size for images,where main structures are over-smoothed or textures cannot be smoothed.Thus,this paper proposes an adaptive scale texture smoothing method.Firstly,it analyzes the statistics in the local rectangular region and selects the appropriate rectangular patch size from the candidate parameter set for each pixel,where in the texture regions larger patch sizes are chosen and smaller ones for regions near the structure edges.Secondly,guided image are computed via the adaptive patch sizes.Finally,the guided bilateral filtering is performed.Experimental results show that the adaptive scale method can achieve desirable results.(2)Most of the existing smoothing methods are based on rectangular patches,which do not coincide with the sharp structures,so the statistical features of pixels within rectangular patches are inaccurate,causing structure edges to be blurred.Thus,this paper introduces an adaptive region texture image smoothing method.Firstly,structure adaptive patches are generated by several times of simple linear iterative clustering segmentation which separate texture from main structures of the image.Then,for each pixel in adaptive regions,the Gaussian weight is used to compute a guidance image and a guided bilateral filtering is performed.Compared with other texture smoothing methods,the approach of adaptive region can obtain better smoothing results.(3)The original random walk segmentation is based on color contrast,while texture images often contain high contrast,so the segmentation results deviate from the image boundaries.This paper proposes a guided random walk segmentation method on texture images.For each pixel,the rectangular patch shift scheme is used to select the patch with smallest tonal range from the patches containing the pixel.The result of the median filtering is used as a guided image in this patch which is used to calculate the weight of random walk segmentation.Finally,image segmentation results of random walk are obtained with the greatest probability.Comparing with the traditional random walk segmentation,this method can achieve more desirable segmentation results.