Makeup transfer based on non-facial area transfer suppression using color channel information

Applying makeup is one method to enhance one's appearance and improve facial impressions. In particular, there has been a growing emphasis on appearance due to the widespread influence of images and videos on social network services. Makeup products and techniques are diverse; therefore, identifying products that meet individual requirements can be challenging. The application of makeup transfer techniques facilitates the easy identification of suitable makeup methods. This study focuses on local adversarial disentangling network (LADN), which pertains to one application of makeup transfer techniques. We identify a problem associated with the LADN, where the transferred makeup images include elements beyond makeup. We aim to enhance the performance of the LADN by suppressing non-makeup elements through the implementation of a proposed loss function. As a result, the proposed method effectively suppresses non-makeup transfers exceeds the performance of the original LADN through evaluation experiments.