1997 Volume 81 Issue 2 Pages 73-79
A new color quantization algorithm is proposed in which skin color data of an image is quantized keeping the original color reproduction as much as possible, while non-skin color data is quantized with the most data reduction.
This study found that effective image quantization is accomplished with the equidivision procedure of cumulative distribution of image data in L*H*C* space (subspace of L*u*v* space) because the optimal weighting process between high and low frequency data for image data distribution can be achieved. However, using this procedure the color reproduction of an quantized image changes with the occupancy ratio of skin color in the image.
Subjective discrimination of the difference in color reproduction between a sample image and the original one varies according to the presence or absence of skin color in the image.