Malaysian Ethnicity Classification based on Face Images using Deep Learning

A face processing system will bring more benefits to numerous applications in surveillance systems, and image or video analysis if the ethnicity information is considered as part of the input data. This will also bring a challenge in face processing studies. Ethnicity information is one of the human characteristics that play a critical role in biometric recognition. Existing face processing approaches usually include two stages: (i) collecting features from face images; and (ii) using the extracted features as the input in a classifier. This paper tackles ethnicity classification based on face images by utilizing a deep learning model. In this project, we address the problem by extracting features and classifying them concurrently using the Convolutional Neural Network (CNN). The proposed method evaluates the classification of Malaysia ethnicity: Malay, Chinese, and Indian. Our dataset is comprised of Google Images and profile images collected from selected participants. The dataset is then annotated with the ethnic group information based on somatic facial features which a human use to distinguish the ethnicity categories. An average of 70% prediction accuracy is reported