Spectral Analysis of Masked Signals in the Context of Image Inpainting

This paper proposes a computationally efficient algorithm for evaluating a sum of squared differences in the image domain in the presence of arbitrary mask configurations. Among the many potential applications of this algorithm, we consider for illustration an image inpainting task. The results show that on a diverse sample of hundreds of simulated holes in the tested images, the proposed technique is more effective than the baseline normalized cross-correlation, even when the masks are properly dealt with by the baseline.