Region Adaptive Single-image Super-resolution Using Wavelet Transform

Abstract

An efficient single-image super-resolution based on the discrete wavelet transform is proposed. The low-resolution image to be enhanced is assumed to be the low-frequency subband of the high-resolution image to be reconstructed. A support vector machine that synthesizes the high-frequency subband based on the inter-subband correlation of the directional edges is designed. Subband coefficients are classified into one of homogeneous region, edge region, and textured region. The support vector machine is adaptively trained for each region. Experimental results of sample images show that the proposed system offers improvements in terms of both measured distortion and subjective appearance.