|
Abstract:
|
Color in any pixel of a digital image is represented in the RGB model in terms of combinations of the three primary colors, namely red, blue and green. Alternative models that provide equivalent representations are the tristimulus XYZ, LUV, HSB/HSL (hue, saturation, brightness/lightness), HCL (hue, chroma, luminosity) colorpaces and so on. These representations of color are bijective from one colorspace to the other, however being nonlinearly related, each representation is not uniform in terms of perception. We investigate the performance of k-means color quantization algorithms in each of these spaces and provide an adaptive approach for the best (in terms of the widely-used peak-to-signal noise ratio) lossy representation of such images.
|
