Smoothed Local Histogram Filters

摘要

Local image histograms contain a great deal of information usefulrnfor applications in computer graphics, computer vision and computationalrnphotography. Making use of that information has beenrnchallenging because of the expense of computing histogram propertiesrnover large neighborhoods. Efficient algorithms exist for somernspecific computations like the bilateral filter, but not others. Here,rnwe present an efficient and practical method for computing accuraternderivatives and integrals of locally-weighted histograms overrnlarge neighborhoods. The method allows us to compute the location,rnheight, width and integral of all local histogram modes at interactivernrates. Among other things, it enables the first constanttimernisotropic median filter, robust isotropic image morphologyrnoperators, an efficient “dominant mode” filter and a non-iterativernalternative to the mean shift. In addition, we present a methodrnto combat the over-sharpening that is typical of histogram-basedrnedge-preserving smoothing. This post-processing step should makernhistogram-based filters not only fast and efficient, but also suitablernfor a variety of new applications.