Improved autoregressive image model estimation for directional image interpolation

Xiong, Ruiqin,Wenpeng Ding,Ma, Siwei,Wen Gao

DOI: https://doi.org/10.1109/PCS.2010.5702531

2010-01-01

Abstract:For image interpolation algorithms employing autoregressive models, a mechanism is required to estimate the model parameters piecewisely and accurately so that local structures of image can be exploited efficiently. This paper proposes a new strategy for better estimating the model. Different from conventional schemes which build the model solely upon the co-variance matrix of low-resolution image, the proposed strategy utilizes the covariance matrix of high-resolution image itself, with missing pixels properly initialized. To make the estimation robust, we adopt a general solution which exploits the covariance matrices of both scales. Experimental results demonstrate that the proposed strategy improves model estimation and the interpolation performance remarkably.