Scale-Adaptive Bilateral Texture Filtering
Song Chengfang;Xiao Chunxia;School of Computer Science, Wuhan University;
Extracting meaningful structures from images with complicated texture patterns is still challenging, since it is hard to separate structures from texture of multiple scales. In this paper, we propose a scale-adaptive structure-aware texture filtering algorithm to smooth out texture while preserving dominant structures. We propose a texture filter that exploits various kernels of multiple scales in one filtering pass, namely, multi-scale texture filtering. The kernel scale for a pixel is evaluated based on its maximum neighborhood clear of structures and features, thus, smaller kernels for pixels on structure edges and bigger kernels for pixels in textural regions respectively. Our method outperforms previous methods in terms of effectiveness and efficiency of texture filtering, as well as preserving salient structures and small-but-important features. Finally, the proposed method achieves commendable experimental results at less cost of computation.