Multiframe image superresolution based on cepstral analysis

Abstract

In this paper, multiframe image superresolution algorithm based on cepstral analysis has been proposed. The multiple low-resolution images are registered by calculating the translational, rotational and scaling shifts in the cepstral domain. The registered multiple low-resolution images are projected onto a high-resolution grid to generate a new high-resolution image. The projection of low-resolution image causes blurred and jaggy edges in the high-resolution image. To de-blur and regularize high-resolution image, Wiener filter is designed using point spread function computed with the help of the shaken length N and shaken angle θ that are obtained from cepstral analysis. Performance of the proposed algorithm is evaluated using different benchmark images in terms of subjective results and objective parameters such as PSNR and computation time. The subjective and objective results show the superiority of the proposed algorithm over the conventional and state-of-the-art superresolution techniques.