Image interpolation based on inter-scale dependency in wavelet domain

Abstract

Image interpolation in the wavelet domain can be considered as the estimation of wavelet coefficients in the highest frequency subband. In this paper, a novel image interpolation method based on inter-scale dependency in the wavelet domain is proposed. In our method, the Gaussian mixture model (GMM) is used to estimate the magnitude of the wavelet coefficient, and the parameters of the GMM are derived from subbands with no training. The sign of the estimated wavelet coefficient is also obtained by using the inter-scale dependency of wavelet subbands. In the simulation results, the proposed method shows an improved PSNR and subjective quality compared with conventional bicubic method and the statistical method (K. Kinebuchi et al., May, 2001) that exploits the hidden Markov tree (HMT) model with training.