Feature space reduction using multicollinearity features

New method for reduction of the feature space in pattern recognition is discussed. The main idea is elimination of the most spurious feature vector components. As the criterion for such informative distinction new feature called diagonal prevalence index is proposed. The efficiency in a sense of both discrimination ability and computational complexity is being discussed in comparison with other multicollinearity features. Finally, we list algorithm for the image recognition based on the diagonal prevalence idea implementation.