Color Style Transfer Techniques using Hue, Lightness and Saturation Histogram Matching

Abstract

We present new methods which transfer the color style of a source image into an arbitrary given target image having a different 3D color distribution. The color transfer has a high importance ensuring a wide area of applications from artistic transformation of the color atmosphere of images until different scientific visualizations using special gamut mappings. Our technique use a permissive, or optionally strict, 3D histogram matching, similarly to the sampling of multivariable functions applying a sequential chain of conditional probability density functions. We work by order of hue, hue dependent lightness and from both dependent saturation histograms of source and target images, respectively. We apply different histogram transformations, like smoothing or contrast limitation, in order to avoid some unexpected high gradients and other artifacts. Furthermore, we use dominance suppression optionally, by applying sub-linear functions for the histograms in order to get well balanced color distributions, or an overall appearance reflecting the memory color distribution better. Forward and inverse operations on the corresponding cumulative histograms ensure a continuous mapping of perceptual attributes correlating to limited derivatives. Sampling an appropriate fraction of the pixels and using perceptually accurate and continuous histograms with minimal size as well as other gems make this method robust and fast.