Confidence-aware guided image filter

The Guided Image Filter has recently been developed as an edge-preserving smoothing filter. It has proven to be effective in a wide range of applications, including noise reduction, haze removal, image matting, and others. In this paper, we propose an extension of the Guided Image Filter that takes advantage of confidence information associated with the filter input. Sometimes, such confidence information is generated in a preceding estimation stage in the processing pipeline, but is not considered by the classical Guided Image Filter. Taking the example of alpha matting, we show that our confidence aware Guided Image Filter can significantly increase the quality of the filter output with small computational overhead, and under maintenance of its O(N) time complexity.