The available color space transform (CST) algorithms in image processing pipelines operate suboptimally. The linear (matrix) color coordinate transformation model can theoretically work 1.5–2 times better than the current approach used in modern cameras. Study on the structure of the CST matrix space has shown its low dimensionality: the Hausdorff dimension is about 3 and the effective PCA dimension is about 4. Based on this analysis, new parameterizations for the transformation matrices have been proposed. It has been experimentally confirmed that the proposed parameterizations improve the transform quality by 8.7% as compared with the available methods.
Ivanov et al. (Fri,) studied this question.