Hybrid Wavelet and Discrete Cosine Transform Methods for Ethnicity Identification

Abstract

Ethnic (Race) denotes to people who share common facial features that perceptually discriminate them from members of other ethnic groups. Ethnicity identification from face images is a process of facial features compilation of an individual compared to existing faces to inference his/her ethnic class, it play important role in face-related applications. In this paper, an improved ethnicity identification system’s accuracy is proposed through hybrid Wavelet and DCT (Discrete Cosine Transform) global feature extractor. Firstly, a wavelet transform is applied to the face image with 4 levels of decomposition. And then the DCT transform is performed on the LL4 (Low-Low) approximation coefficients band. A frontal facial database containing (950) images of three different ethnic groups (European, Oriental, and African) with different conditions (lighting, expression, with glasses or without) is used for experimental tests. Results show that the proposed scheme is outperform other recent related works in terms of accuracy and efficiency.