The memory colour reproduction is an important factor in judging image quality of photographic images of real life scenes. As the most important memory colour category, skin tone was extensively studied for preferred colour reproduction in this research. The methodology to study skin colour preference was then applied to study the colour preference of two other important colour categories: green foliage and blue sky. There are three essential parts for preferred skin colour enhancement: 1) building a skin colour model to detect skin colours or skin pixels; 2) finding a preferred skin colour region or a preferred skin colour centre; and 3) developing an algorithm to morph skin colours toward the preferred skin colour region. This study for skin colour enhancement started with the mathematical modelling of the skin colour region for skin colour detection. The modelling of skin colours was then applied to adjust skin colours of test images for psychophysical experiments that were to determine a preferred skin colour region. Finally, the skin colour modelling and the preferred skin colour centres were applied to adjust skin colours of digital photographic images for preferred colour reproduction. Two approaches were developed to model the skin colour distribution for skin colour detection. The first approach was to model a local colour region for general applications. A convex hull is constructed to fit the geometrical shape of a local region, and then the convex hull is approximated with mathematical formulae. The formulations and data fitting are adjusted with interactive 3-D visualization. The approach is flexible for fitting data gamut with various mathematical forms for different purposes. The other approach was to model skin colours with elliptical shapes. Three elliptical skin colour models were developed for skin colour detection. The first one is to model the skin colour cluster using a single ellipse ignoring the lightness (or luminance) dependency. It is simple and efficient, and the skin colour detection accuracy may be adequate for many applications. In the second model, the skin colour ellipse is adapted to different lightness so that the shape of the ellipse fits the skin colour cluster more accurately. The model is more complex to train and is less efficient in computation, but it is more accurate in skin colour detection. In the third method, an ellipsoid is trained to fit the skin colour cluster. It is almost as simple to train as the first model, but the skin colour detection accuracy is improved. Finally, these models were applied to train mixed skin colours, African skin colours, Caucasian skin colours, and Asian skin colours.
展开▼
