Extracting meaningful structures and removing unimportant details or textures from a real world image is a classical problem in the domain of non-photorealistic rendering for images and benefits further computer vision related tasks as well. In this paper, we propose a structure sensitive bilateral filter by designing a novel weighting function which applies the second order variations of local color values. The proposed filter results in an image stylization effect where an image is represented by more flatten while distinguishable color regions. Furthermore, by combining the traditional bilateral filter as a post processing, textures embedded in mosaic art and embroidery art like images can be removed efficiently. Comparisons with the state-of-the-art methods are also provided to verify the validity and efficiency of the proposed filter.