Abstract: Texture transfer is a kind of stylization techniques for images, which transfers texture information from an example image to the source image. However, existing methods cannot work well when the source image contains rich textures or the example image is complex, and their interaction is unintuitive. In this paper, we propose a new texture transfer method based on non-linear filtering. Firstly, textures of the source image are removed using relative total variation（RTV） filtering and L₀ filtering. Secondly, textures of the example image are extracted using RTV filtering. And finally, the extracted textures are transferred to the source image by illustration remapping and linear superposition. Experimental results show that the proposed method can preserve the structure and tone information of the source image while update its textures by those from the example image. Compared with existing methods, our results are more favorable and our interaction is more intuitive.