Quaternion-based color image smoothing using a spatially varying kernel

Abstract

Addressing the issue of feature/detail preserving color image smoothing, we propose a novel unified approach based on a quaternion framework. The main idea is to holistically extract the local orientation information at each lattice point, and then to incorporate it into the smoothing process. We introduce a new Quaternion Gabor Filter to derive the local orientation information in color images. This derived orientation information is modeled using a continuous mixture of appropriate exponential basis functions. We solve the continuous mixture integral in analytic form, and develop a spatially varying kernel which respects to the local geometry at each lattice point in a color image. Superior performance of our smoothing framework is demonstrated via comparison to competing state-of-the-art algorithms in literature. © 2009 Springer.