The appearance and color distribution of skin are important characteristics that affect the human perception of health and vitality. Dermatologists and other skin researchers often use color and appearance to diagnose skin conditions and monitor the efficacy of procedures and treatments. Historically, most skin color and chromophore measurements have been performed using reflectance spectrometers and colorimeters. These devices acquire a single measurement over an integrated area defined by an aperture, and are therefore poorly suited to measure the color of pigmented lesions or other blemishes. Measurements of spots smaller than the aperture will be washed out with background, and spots that are larger may not be adequately sampled unless the blemish is homogenous.
Recently, multispectral imaging devices have become available for skin imaging. These devices are designed to image regions of skin and provide information about the levels of endogenous chromophores present in the image field of view. This data is presented as four images at each measurement site including RGB color, melanin, collagen, and blood images. We developed a robust segmentation technique that can segment skin blemishes in these images and provide more precise values of melanin, blood, and collagen by only analyzing the segmented region of interest. Results from hundreds of skin images show this to be a robust automated segmentation technique over a range of skin tones and shades.