A Gradient-Based Adaptive Interpolation Filter for Multiple View Synthesis

Ping Yang,Xin Tong,Xiaozhen Zheng,Jianhua Zheng,Yun He

DOI: https://doi.org/10.1007/978-3-642-10467-1_49

2009-01-01

Abstract:A gradient-based adaptive interpolation method is proposed in this paper to solve the over-blurring problem in conventional multiple view synthesis (MVS) filters. To improve the visual quality of final synthetic pictures, a good interpolation filter is required in multiple view synthesis steps. Traditional space-invariant filters, such as bi-linear or bi-cubic filter, take the advantage of complexity, but they also lead to a decrease of the subjective quality. In contrast, directional filters usually exploit the directional information, especially in edge area, to deal with the over-blurring problem. This paper proposes a fast directional interpolation method for scaling up the resolution or filling up the losing pixels of a picture. The gradient map of an input picture is calculated in first. Using the gradient of each input pixel, the interpolation coefficients computed from a Gaussian kernel is refined, leading to a directional filter which takes an adaptation with gradient direction. This method is with a low complexity because it is a non-iterative method, and experiment results show that the visual quality of interpolated picture is improved.