Abstract

We propose the propagation fifilter as a novel image fifiltering operator, with the goal of smoothing over neighboring image pixels while preserving image context like edges or textural regions. In particular, our fifilter does not to utilize explicit spatial kernel functions as bilateral and guided fifilters do. We will show that our propagation fifilter can be viewed as a robust estimator, which minimizes the expected difference between the fifiltered and desirable image outputs. We will also relate propagation fifiltering to belief propagation, and suggest techniques if further speedup of the fifiltering process is necessary. In our experiments, we apply our propagation fifilter to a variety of applications such as image denoising, smoothing, fusion, and high-dynamic-range (HDR) compression. We will show that improved performance over existing image fifilters can be achieved