Object Removal using Recursive Median Filter
Rajkumar Laxmikanth Biradar, A Vinayadatt V Kohir · 2013
In this paper, an object removal in an image is carried out by diffusing surrounding information. The challenge is to fill in the “target region”, hole left behind by the removal of the object and create the homogeneous background in its place. Object removal may be treated as one of the applications of image inpainting. We use recursive median filter and Gaussian pyramid. Gaussian pyramid fills in small target regions and the recursive median filter diffuses the neighboring pixels. By combining Gaussian pyramid and recursive median filter, we achieve removal of larger object. The main idea in the present work is to down sample the input image using Gaussian pyramid and using recursive median filtering the median value of pixels are diffused from the exterior area of the target region into the target region at the lowest resolution. Then we substitute the target region of the image in the second lowest resolution by expanding diffused image at the lowest resolution level. The resulting image is again expanded and the target region in third lowest resolution is substituted. The process of expansion and substitution is carried out till zeroth i.e. original resolution to obtain output image without the object. The down sampling reduces the size of object to be removed along the Gaussian pyramid. At the lowest resolution recursive median filter removes the unwanted object by diffusion. Expansion (upsampling) and substitution constructs the homogeneous background in place of the object. This technique produces better results for removal of medium and large size objects.