Study On Scale-aware Texture Suppression For Structure Extraction And Its Application

Image structure extraction has been long studied in area of digital image processing and computer vision. It is a fundamental problem for image segmentation, object detection, video tracking and many other applications. Natural scenes contains many texture information, both regular, near-regular or irregular. Texture structure and other main structure like contours always mixed with each other, we can easily see a structure over texture areas. We proposed a novel method for extract image structure from noisily background with abundant textures. The structure extracted is texture-scale aware since we can controllable reject some texture structures efficiently as we adjust the texture scale parameter. We compare our method with other structure extraction method and also nearly proposed texture smoothing algorithms. Both result and time efficiency show our advantage over these other methods.Lastly, we find some applications to verify our algorithm in practical. Thus, Japanese manga structure extraction and structure-based ASCII art auto-conversion.