Automated Face Detection Using Skin Color Segmentation and Viola-Jones Algorithm
Viola-Jones algorithm can be categorized as one of an established and effective method (feature-based approach) for detecting face. It consists of three main processes which are Haar features, Adaboost and Cascading. These processes involved scanning the patterns of human face through all the pixels in the image. In the empirical experiment by using this algorithm, some region in the image which is supposed to be non-face region is detected as face due to the similarity of human face features. In addition, this approach only search for pattern (feature) but leaving the color information. Therefore, this paper proposes hybridization of skin color segmentation in prior to Viola-Jones algorithm. Several images with different kinds of environment (different light conditions, different face orientation) and various people ethnics (Malay, Chinese, Indian, etc.) are tested using the improved algorithm and the original Viola-Jones algorithm as well. In average, experimental results reveal the combination of YCbCr color and Viola-Jones algorithm is the best model (average accuracy is ~88%) to detect human face in various conditions. Other color model such as HSV and benchmarked algorithm are having slightly low detection rates due to some false face detection. Despite of this, this research also found that both YCbCr and HSV color model are having some limitation when dealing with darker face and lightning condition since the skin color become slightly out of range from normal skin color distribution.
Copyright (c) 2019 UMP Press
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.