Sufficient Image Appearance Transfer Combining Color and Texture

• Published in: IEEE transactions on multimedia Vol. 19; no. 4; pp. 702 - 711
• Main Authors: Song, Zhi-Chao, Liu, Shi-Guang
• Format: Journal Article
• Language: English
• Published: IEEE 01-04-2017
• Subjects: Algorithm design and analysis; Appearance transfer; color; Feature extraction; Image color analysis; image editing; Lighting; Mood; point set expansion; Prediction algorithms; texture; Visualization
• DOI: 10.1109/TMM.2016.2631123
• EISSN: 1941-0077
• ISSN: 1520-9210

Traditional color transfer methods can achieve satisfactory results for transferring the color style from a reference image to a source image, provided that the source image shares the similar color mood with the reference image. However, color transfer solutions are always sensitive to color category, which cannot generate natural results when the contents of the reference image and the source image are different, e.g., a lush tree in the reference image and a bare tree in the source image. In this situation, it is insufficient only through color transfer to transfer the appearance from the reference image to the source image only through color transfer, since other information such as texture should also be considered. To obtain sufficient appearance transfer results, we propose a new image appearance transfer method combining both color and texture features. Given a source image and a reference image, our method starts with feature detection and matching between the source image and the reference image. Then, we design a new method for expanding feature point sets to get texture transfer mark (TTM) and color transfer mark (CTM). TTM and CTM will guide texture transfer and color transfer, respectively. We demonstrate our appearance transfer algorithm between quantities of images and compare with results of existing methods. Experiment results show that given only a single reference image, our approach can produce more sufficient appearance transfer results than the state-of-the-art algorithms.