Research On Contant-time Image Texture Filter And Its Applications

Texture filtering is an image manipulation that separates salient structures from complex textures.The algorithm is quite useful in a variety of applications.This paper presents a nonlinear texture filtering algorithm based on a variant of weighted median filtering.The guidance image for the weighted median filtering is constructed through the interpolation of the input image and the blurred input image.Since the guidance image incorporates both of the salient structures and flattened textures,the proposed algorithm can remove textures while preserving edges well.Therefore,the texture filter can separate salient structure from texture information.Similar to the constant-time weighted median filter which uses the histogram volume to compute median values,we develops a constant-time algorithm by adopting the histogram volume to calculate local extreme values.Moreover,the local nature of algorithm enabled an efficient GPU parallel implementation.The GPU-based algorithm can improve the processing speed especially on high-end NVIDIA graphics cards.To demonstrate the efficient of our algorithm on texture filtering,we compared our algorithm with state-of-art texture filtering methods.Many applications including detail enhancement,JPEG artifact removal,image segmentation,image stylization,inverse half-toning,and stereo image disparity matching are presented in this paper.These applications demonstrate the effectiveness of the proposed texture filter.