Fast image filter based on adaptive-weight and joint-histogram algorithm

Abstract

Adaptive-weight operators are ubiquitous in numerous computer vision applications. The structure of general adaptive-weight models, however, are hard to accelerate with high speed to large or complex images. In this paper, the proposed adaptive-weight image filter algorithm is mainly on a new joint-histogram representation, median value searching, and a new data structure that contributes to fast data access. The effectiveness of these schemes is demonstrated on estimation of median position, which not only better preserves edges, but also reduces computation complexity from O(mnr2) to O(mnr) using histogram, where m * n and r denote image size and radius of the mask window respectively. The results of our experiments demonstrate that our approach is effective to image filtering and image enhancement.