A novel single-image super-resolution algorithm based on self-similarity in wavelet domain

Single-image super-resolution is a fundamental problem in image processing. This paper proposes a novel superresolution algorithm based on self-similar structure existing in wavelet domain, which utilizes the non-local similarities prior of natural images. In our method, we first establish the relationship connecting different areas in adjacent subbands by fractal encoding in wavelet domain, then extend this relationship and use it to recover the unknown coefficients in the zeroth subband via super-resolution fractal decoding. The desired high-resolution (HR) output image is then obtained by inverse wavelet transform. Experimental results suggest that the proposed method achieves significant improvement in terms of PSNR and subjective visual quality in contrast to most of other super-resolution algorithms, such as bicubic and sparse representation-based methods.