A novel automatic color transfer algorithm between images

Abstract

A color transfer algorithm between images intends to modify the colors in the source image by borrowing the color characteristics from the target image. In this paper, we present a novel automatic color transfer algorithm between images. This algorithm is able to proceed on color transfer with no user intervention, and produce visually plausible resultant images. To the best of our knowledge, our algorithm is the first to achieve this kind of automation. Our technique consists of an image segmentation algorithm, a swatch merging algorithm, and a pattern recognition algorithm. The image segmentation algorithm generates a number of swatches with different features in the source and target images. The swatch merging algorithm merges a particular swatch with its neighbors in the source image when it fails to pass the normality test. The pattern recognition algorithm then automatically matches these swatches for color transfer. We develop a boundary transition algorithm to make possible a consistent transition between two boundaries of patches. Experimental results demonstrate that, without any user intervention, our technique performs color transfer results with good visual realism. The technique is superior to the original user-intervention color transfer algorithm.