Facial analytics is a derivative of face-based soft-biometrics that allows a machine to generate information about a person automatically, extracting this information from their face. Facial analytic applications generate descriptive information about an individual’s face, but does not attempt to identify the subject.The attributes generated from these systems may include face shape, face pose, age, sex and other identifiable information. This work examined different types of texture descriptors, to determine the best descriptor for determination of sex (gender) and race. Further, this work will compared three general purpose machine learning classification techniques. The feature descriptors examined were local binary patterns (LBP), histogram of oriented gradients (HOG) and GIST. Each feature descriptor was married to all of the following classifiers: Artificial Neural Networks, Support Vector Machines and Extreme Learning Machines to determine which combination of feature and classifier generates the best recognition performance for gender and race recognition on two well-researched face databases and one extremely large scale database that contains 1.2 million faces.

Recommended Citation: Raynor J., Ricanek K., Reinicke B., Patterson L., (2013). An Evaluation of Classification Performance for Facial Analytics on Diverse Datasets. UNCW MS CSIS Proceedings. V. 7 , N. 2 .