Image Filtering Driven by Level Curves

Abstract

Abstract. This paper presents an approach to image filtering that is driven by the properties of the iso-valued level curves of the image and their relationship with one another. We explore the relationship of our algorithm to existing probabilistically driven filtering methods such as those based on kernel density estimation, local-mode finding and meanshift. Extensive experimental results on filtering gray-scale images, color images, gray-scale video and chromaticity fields are presented. In contrast to existing probabilistic methods, in our approach, the selection of the parameter that prevents diffusion across the edge is robustly decoupled from the smoothing of the density itself. Furthermore, our method is observed to produce better filtering results for the same settings of parameters for the filter window size and the edge definition.

Automated Filter Parameter Selection Using Measures of Noiseness

“…More recently, new approaches based on spatially-varying regressions [20] and spatially varying convolutions [19] have also been developed. An approach to filtering using level curves in a mean-shift framework has been developed in [17].…”