Virtual Makeup Based on Golden Sample Search

Abstract

Virtual makeup is expected to provide customers with an approximate experience of actual trial. This paper proposes a novel virtual makeup system based on golden-sample search scheme. A golden-sample database is built offline, which contains canonical patterns of facial components to be made-up. A novel feature representation of the pattern is developed, which makes the golden-sample search effective and efficient. In online phase, users select the markup region and colors they like. The representative feature vector of the selected markup region is used to search the database to find the best-matching pattern. Then, the best-matching pattern is colored and stitched to the input image. The system can process the dynamic image from camera, and thus the user can observe whether the makeup effect under different facial poses is appropriate, and then make corresponding makeup adjustments. The experimental results indicate our system provides nature and good visual effects in makeup.