The three main factors determining the camera output are the illumination SPD, the camera"s channel sensitivities and the reflectance of the object (for more details, see Section 3.3). Another important factor is white balancing or white calibration conditions which are typically done so that a white object will have a white colour appearance. Even proper calibration does not guarantee that colours other than white (and the other achromatic colours if the camera is linear) will appear the same between images taken under different white balancing conditions. Moreover, when the illumination used in white balance is not the same as the prevailing illumination, the appearance of the colours will most likely change in a more drastic way. The bigger the colour temperature difference between these two illuminations, the bigger the shift of chromaticities, which is eventually limited by the properties of the camera. Also the direction of colour temperature change matters. When the prevailing illumination has a higher colour temperature, it makes colours appear more blueish, and when the opposite happens then the colours appear reddish (see Section 4.2).
In practical situations, there can also be other factors causing colour shifting. Nonlinearity of the camera is one factor and it is caused by the input-to-output conversion in which the transformation and its parameters depend on the input signal. The nonlinearity can manifest itself as a changing slope between the input and the output or gamma factor. There are many cameras in which the nonlinearity is deliberately introduced to compensate for nonlinearity in the display (Holst 1998 and Klette et al. 1998). In some cameras, the nonlinearity is introduced only in the high and low intensity areas. As an example of this kind of nonlinearity, the Sony DXC-755P has a knee-phenomena for the protection of the CCD elements from intense light (Sony 1989). In the knee-phenomenon, when the intensity of the light is higher than a certain threshold, the slope will become smaller. Another effecting factor is the limited dynamic range of the camera. This manifests itself in clipping of values: overclipping if the values are saturated to the maximum value and underclipping if they are zeros. Both forms of clipping cause serious information loss and in many algorithms pixels with clipping are ignored or it is assumed that there is no clipping in the image. Also the camera settings and controls, background, smearing etc. have their own influence on the formation of the image.
The purpose of a skin locus is to define an area of possible skin chromaticities perceived by a colour camera under a certain illumination variation range. It should be emphasized that no specific distribution is assumed for the skin colour or skin chromaticities. In this thesis, the chromatic constraint created does not only offer robustness against illumination change, but also independence of the camera white balancing conditions because it consists of subloci obtained from different camera calibration conditions (Paper IV). In any case, it is camera dependent or more accurately, sensor dependent.
Two methods will be presented for gathering this information about skin colour change. The goal of both methods is to find a good model for the chromatic constraint. It will also be shown that the NCC rgb space is an appropriate colour space for the constraint and therefore it is used in both methods. However, there is no reason why some other colour space might not also be applicable for these methods, although the quality of the chromaticity constraint created in these spaces cannot be guaranteed. The first of the methods presented here uses images with skin to create the chromatic constraint (skin locus). The images are taken under such illumination conditions which are thought to be representative to those one encountered in the application. For the second method, the spectral data of the camera, illumination and object need to be known or measured.