5.2. Skin locus from an image series

The chromaticity constraint (called skin locus in NCC space) is formed from a set of images using the method presented in Section 4.2. This kind of image set is shown in Figure 17 taken by a Nogatech camera (Paper IV). For this image series the selected illumination range varied from the sunset / sunrise conditions to daylight condition. The selection is based on the evaluated robustness needed against the possible illumination changes encountered in the application. It is usually enough to take illuminants from both extremes and a couple from the middle range with suitable colour temperature differences. However, one should be cautious with fluorescent light sources. If a fluorescent light source with strong green components is not included in the locus, then the skin chromaticities perceived under this light source might fall outside the locus. With fluorescent lamps which have a strong red and / or blue component, this is not such a big concern because variations caused by them are usually smaller than those caused by colour temperature changes.

The range of illumination and white balancing conditions in which a skin locus is a valid description of possible skin chromaticities effects its size. If only one white balancing is used, then the obtained locus is smaller than the one which was constructed using many white balancing conditions. The former locus is more specific being only valid for one camera calibration. The later locus is bigger and offers robustness against calibration conditions at the expense of reduced discrimination capabilities between skin and non-skin objects. Therefore, when before making the locus one should consider both the illumination range encountered in the application and the possible issues related to white balancing of the camera.

After the series has been obtained, the skin areas in the images are manually selected and then extracted as shown in Figure 18 for the Nogatech camera.

Then the extracted skin pixels are converted from RGB space to NCC rbgI space or to some other colour space. In Figure 19, NCC gb coordinates of skin pixels are visualized. It is possible to model the obtained skin locus with functions describing the boundary area or with a look-up table.