Grayscale aerial and space image colorization using texture classification

This paper introduces a new approach to colorization of grayscale aerial or space imagery. Proposed method is based on a simple premise: similar textures should have similar color distributions. This premise is formalized by using Bayesian texture classification of grayscale destination imagery with a set of textural prototypes or swatches from source color imagery. Only 2D distributions of chrominance channels are transferred and original luminance value is retained. A special segmentation technique is proposed in order to increase the performance of the algorithm. Examples on real images show high quality of colorization neither without using low-resolution color imagery nor without any user intervention. Appropriate prototype selection enables modeling of season changes of vegetation. It is shown that the proposed approach can be successfully applied to color infrared (CIR) imagery transform into visible RGB colors and for color image remapping.