Improved single image super-resolution using sparsity and structured dictionary learning in wavelet domain

This paper introduces a single-image superresolution approach which is based on sparse representation over dictionaries learned in the wavelet domain. The diagonal detail subband learning and reconstruction is improved by designing two diagonal dictionaries; one for the diagonal and another for the anti-diagonal orientations. Four pairs (low resolution and high resolution) of subband dictionaries are designed. The sparse representation coefficients for the respective low and high resolution images are assumed to be the same. The proposed algorithm is compared with the leading super-resolution techniques and is shown to excel both visually and quantitatively, with an average PSNR raise of 0.82 dB over the Kodak set. Moreover, this algorithm is shown to significantly reduce the dictionary learning computational complexity by designing compactly sized structural dictionaries.