Evaluation of Uniform Colour Spaces for Image Processing

Details:

Year: 1994
Pages: 20

Summary:

Since the adoption of CIELAB and CIELUV in 1976, several other uniform spaces have been developed. We evaluated most of them by making use of visual data and found that they could be divided into two different classes, i.e. spaces to predict small colour differences and spaces to calculate larger differences. We indicated the main difference between both classes and constructed a simple Colour Difference Formula (CDF) for small colour differences. In contrast to grey image processing techniques, almost no attention has been paid to colour image processing. Therefore we determined some basic rules to adapt grey image processing procedures for colour images in such a way that the images are handled in agreement with the human visual system. This was mainly done by making use of tristimulus spaces and uniform colour models. However, as opposed to what was expected, it is almost impossible to detect differences between most uniform spaces even if they differ significantly according to visual data. We indicated that this was mainly due to non uniformities of the colour spaces, non modelled visual effects and quantization errors. In contrast to grey image processing techniques, almost no attention has been paid to colour image processing. Therefore we determined some basic rules to adapt grey image processing procedures for colour images in such a way that the images are handled in agreement with the human visual system. This was mainly done by making use of tristimulus spaces and uniform colour models. However, as opposed to what was expected, it is almost impossible to detect differences between most uniform spaces even if they differ significantly according to visual data. We indicated that this was mainly due to non uniformities of the colour spaces, non modelled visual effects and quantization errors. Based on this evaluation we recommend the use of different spaces for the determination of small and larger colour difference. The chosen spaces should be mathematical stable and simple. A good candidate for small resp. larger colour difference is CDF, resp. CIELAB.