Sub-optimal Parameter Estimation for Bilateral Filter using Distribution Distance

Abstract

A method that estimates the parameters for a bilateral filter, based on the distribution distance of the images, is proposed. This study presents a method to estimate a sub-optimal set of parameters in a reasonable execution time by using a local search method with adequate initial values. The image quality is slightly degraded when sub-optimal parameters are used, and therefore a second method is proposed to compensate the parameters to achieve better image quality. Parameter estimation time was reduced from 398 s (brute-force method) to 0.15s (proposed local search method) with a 2.8GHz Xeon processor. When the compensated parameters are used, the image quality was approximately equal to the optimal result.