Abstract

In consumer imaging applications involving photo collages or composition of user photos with professional artwork, inconsistent color appearance of photos and artwork from different sources can result in compositions that do not look aesthetically pleasing.  Users often express a desire to modify individual images to achieve a more consistent color appearance.  Prior work in color transfer that extracted the color properties of one image and applied it to another has shown very interesting results.   Those works focused on achieving an artistic effect, usually without the constraint of conserving object color.   In consumer imaging, we have to be more conscious about conserving general object color and especially skin tones, which are not amenable to aggressive color change.   In this talk I will describe an algorithm to estimate the color and tone properties of an image and transfer these properties to another image under a strong naturalness constraint.  With this method, color changes are constrained to correspond to incomplete adaptation under natural illuminants.  We use a simple Bayesian method to characterize scene color properties, expressed as scene color temperature and illumination levels.  An existing color adaptation model RLAB is used to apply color changes by simulating incomplete adaptation to a target illuminant.  We emphasize that this is not a method of white point estimation nor a white balance  procedure.  Rather, we use color adaptation models as a means to ensure color adjustments to be “plausible”, and therefore maintain a natural appearance to the images even after significant color adjustments.

Speaker

Xuemei Zhang has worked as an imaging scientist/architect at HP Labs, Agilent Labs, Micron Technology, and currently at Apple. She received her BS in psychology from Beijing University and her Masters in statistics and PhD in psychology from Stanford University.  Her work has focused on imaging algorithms for cameras and displays, as well as color image quality metrics.  This talk reviews work on color transfer she did (in collaboration with Hui Chao and Daniel Tretter) while she was a senior research scientist at HP Labs.