Image interpolation based on universal hidden Markov tree model

In this paper, an image interpolation method is proposed based on universal hidden Markov tree (HMT) model. The universal HMT model shows a good score for noise reduction in wavelet domain and furthermore, it needs no training procedure J. K. Romberg et al. (2001). In proposed method, the model is modified properly for image interpolation application as follows. Firstly, the state of parent coefficient is determined by the subband standard deviation. Secondly, the variances for Gaussian mixture model (GMM) of universal HMT model are estimated directly from input image by using the property of exponential decay of variance across subband. In simulation, the proposed algorithm shows an improved performance compared with conventional bicubic method and the method based on HMT model with training procedure by similar images G. D. Haan et al. (1998).