Key points are not available for this paper at this time.
Colours are usually described using three perceptual variables: an achromatic variable, such as lightness or luminance, and two “chromatic” varibles, such as chroma and hue. This means that the conversion of colour images to greyscale is often thought of as the removal of “chromatic information”, which leaves a greyscale made by just the achromatic colour variable; i.e. luminance or lightness. One obvious problem with this approach is how to make a greyscale for equiluminant images, or images containing equiluminant object boundaries.In this paper we review some of the more recent attempts to tackle the colour-to-greyscale problem. We then use an image preference experiment to test the performance of these methods on a selected set of images; some of which lose a significant amount of information when converted to greyscale using luminance, and others which do not. The results show that, in general, the newer techniques can provide a greyscale image which is preferred to that derived by luminance, especially for images that have prominent equiluminant boundaries. The results also show that this advantage is not guaranteed for every image, and that no particular algorithm provides consistently better performance than the others.
Connah et al. (Mon,) studied this question.