Imaging Flesh: Skin-customized Profiling

摘要

Choice of camera calibration target influences how well an algorithm can estimate the color of specific scene objects. This research investigates the value of customized targets for estimation of flesh color. The use of a standard color calibration target is compared with a customized skin-based target. A spectral database of measured skin reflectances was available for use in this study. A skin target was simulated and compared to how well a standard MacBeth ColorChecker performed in calibrating a camera for accurate estimation of skin colors. The ColorChecker had a low average of 2 C1EDE2000 error units for the task. A simulated flesh target improved estimation error by 26%. For a second simulation a noise model derived from a commercial clinical photography system was imposed. In this more realistic case, the skin-customized target improved results by afar larger 60%.