Directional Moving Averaging Interpolation for Texture Mapping

Abstract In texture mapping, several filtering techniques have been proposed which allow prefiltering of a texture. It has been demonstrated that the shape-variant filter kernel has better performance than its shape-invariant counterpart. However, the conventional shape-variant filtering in which the filtering area is warping-based uses the same kernel for high-detail areas as for low-detail areas of the image in texture space. In this paper we propose a hybrid interpolation filtering method for texture mapping which is based on content-based space-variant directional filtering concepts. First, we classify the local image area in the texture space into three types of blocks: constant, oriented, and irregular blocks. Second, for constant and irregular blocks, we use local average filtering and elliptical weighted average (EWA) filtering, respectively. Third, for oriented blocks, we propose a new filtering method called elliptical weighted adaptive directional moving average (EWADMA) filtering. In the experiments, we show that our method may recover the distorted images with better subjective quality.