An analysis of automatic gender detection by first-order configural relations

Automatic gender detection from face images is a challenging problem. In the literature, different techniques have been applied so far on face images for gender detection. In contrast to these existing methods, we have analyzed the usage of first-order configural relations of the face to predict gender from images by using machine learning algorithms. In experiments on the dataset of Wikipedia profile pictures, 83% of general accuracy, 83.3% detection rate for male faces and 82.7% detection rate for female faces have been achieved by Logistic Regression. These results indicate that first-order configural relations are effective in automatic gender prediction from digital face images.