A Method for Obtaining Human Skin Color Pixels Based on Surrounding Factors

Abstract

Accurate skin color detection methods are very valuably used in diverse fields such as face recognition and tracking, facial expression recognition, adult image detection and healthcare. In this paper, the performances of diverse skin color detection algorithms were comparatively evaluated as the distance to the subject was changed and the color of the subject background was changed in the environment where normal light and indoor lighting were added. The test was conducted by selecting 2 males and 1 female having different skin tones as the subjects and dividing the background colors into white, black, orange, pink and yellow. The skin color extraction algorithms used for performance evaluation were Peer algorithm, NNYUV algorithm, NNHSV algorithm, LutYUV algorithm, and Kismet algorithm. The test was conducted by limiting the distance between camera and subject to 60cm to 120cm. As a result of conducting the performance measurement test, the performances of the algorithms showed differences according to the changes made to the subject background. In general, NNHSV and NNYUV algorithms using neural network and LutYUY algorithm showed stable results. As far as the remaining algorithms were concerned, their skin color detection rates were significantly influenced by the changes made to the background. This paper is expected to be effectively utilized in developing an adaptive and highly accurate environment-based skin color extraction algorithm suitable for actual environments where dynamic changes are made to the subject's surrounding environments.