Spatial color imaging: from Retinex to ACE

What gives us the final sensation of colour is not only the colour signal. The appearance of colour in real scenes can vary widely according to several factors. The two major of them are illuminant and context. These two factors seem to “pull” our vision system in two different directions. From one side colour constancy mechanisms make the object colour more stable under the changes of light sources spectral composition while, from the other side the effect of the context makes the object colour depending on the scene spatial composition and consequently less stable in itself. These two apparently contradicting phenomena are based on the same principle: the spatial recomputation of the colour signal. It produces the final overall appearance of the scene content. So far, several algorithms have tried to simulate this visual normalization mechanism. Their two main basic macro behaviours are Gray World and White Patch, which are considered alternatives. Considering them separately, they produce two different normalization mechanisms. Lightness Constancy and Color Constancy. Gray World approach goes in the Lightness Constancy direction: it centers the histogram dynamic, working in the same way as a camera exposure control. White Patch approach goes in the Color Constancy direction, searching for the lightest patch to use as a sort of illuminant reference. Retinex algorithm basically belongs to the White Patch family due to its reset mechanism. Searching a way to merge these two components, we developed a chromatic correction algorithm, called Automatic Color Equalization (ACE), which is based on both. It maintains the main Retinex idea that color sensation derives from a local comparison of the spectral lightness values across the image. We present the common ground of the two algorithms, their differences and their results.