String-Like Occluding Region Extraction for Background Restoration

In this paper, we propose a method for extracting string-like objects in a still image for background restoration. We assume that the object regions occluding the background are long and narrow, and contrasted in intensity with background. First the method introduces a circle contrast, the intensity difference between a pixel and those on a circle around, to find the occluding string-like regions. Then the signs of the circle contrast are decided so that the occluding regions and backgrounds are well separated, and further enhanced by an optimization process. Extracted regions are removed with interpolation (inpainting) for background restoration. Experimental results on real images show the validity of the proposed method