Exemplar-based image inpainting using svd-based approximation matrix and multi-scale analysis

Reconstruction of images by digital inpainting is an active field of research and such algorithms are, in fact, now widely used. In conventional methods, a texture synthesis algorithm is used for filling the unknown regions of the image. However, due to the lack of global analysis on the image, the result may contain undesirable artifacts especially when we have images with relatively large missing regions. Here we propose a new inpainting technique to overcome this limitation by using an approximation matrix. The basic idea is to first make an approximation matrix using singular value decomposition, and then reconstruct the target region by using this matrix. Approximation matrix here, is in fact a gray-scale copy of the original image in which the target region is approximated throughout the process of rank lowering. Experiments are performed on a variety of input images ranging from purely synthetic images to full-color photographs. The results demonstrate the effectiveness of the proposed approach.