Homologous Anatomical-Based Facial-Metrics Application to Down Syndrome Face Recognition.

Down syndrome (DS) is one of the prominent neuro-developmental diseases which are distinguished by atypical fractionation behaviors, physical traits, and other mental disabilities. Current techniques of recognizing the syndrome need genetic testing through clinical studies, which is usually expensive and challenging to get. In order to simplify the classification approach, computer-aided facial analysis methods incorporating machine learning and morphometrics are crucial. Thus, this study proposes Homologous Anatomical-based Histogram of Oriented Gradients plus Support Vector Machine (HAB-HOG/SVM) to automatically detects and extracts 74 homologous facial landmarks from the subjects (DS patient and healthy control) face image and Chord-Transformed Principal Components (CT-PC) as features extraction method for classification. The novelty of this method relies on the automatic acquisition of landmark data which is conceptually simple, robust, computationally efficient, and annotation error-free and the feature extraction technique applies which is simplified enough to follow. The experiment reports recognition accuracy of 56.82% and 98.86% for Classical Principal Components (CPC) and Chord-Transformed PC, respectively. The results demonstrate that the suggested method outperformed not only the CPC but also the previously presented state-of-the-art methods in the domain of DS face recognition.