Color Transfer by Steerable Pyramid

This paper describes a steerable pyramid-based method for color transfer between images. The primary idea of the work is that the statistical information is transferred from the target subbands to their corresponding source subbands respectively, where all those subbands are analyzed by a steerable pyramid. Statistical information in this paper includes expectations, standard variances, skewnesses and kurtoses of all subbands. The pixel values of source image and target image are first transformed into the lαβ color space in order to prevent undesirable intercrosses between R, G and B channels. The transformed channel data are then analyzed respectively by a steerable pyramid, and statistical information is finally transferred during collapsing the pyramid. Being a flexible architecture, steerable pyramid provides a mechanism for selecting the needed subbands to be transferred, which results in generating different results.