Mayo, M. & Zhang, E. (2008). Improving face gender classification by adding deliberately misaligned faces to the training data. In Proceeding of 23rd International Conference Image and Vision Computing New Zealand 2008 (IVCNZ 2008).
Permanent Research Commons link: http://hdl.handle.net/10289/2172
A novel method of face gender classifier construction is proposed and evaluated. Previously, researchers have assumed that a computationally expensive face alignment step (in which the face image is transformed so that facial landmarks such as the eyes, nose, chin, etc, are in uniform locations in the image) is required in order to maximize the accuracy of predictions on new face images. We, however, argue that this step is not necessary, and that machine learning classifiers can be made robust to face misalignments by automatically expanding the training data with examples of faces that have been deliberately misaligned (for example, translated or rotated). To test our hypothesis, we evaluate this automatic training dataset expansion method with two types of image classifier, the first based on weak features such as Local Binary Pattern histograms, and the second based on SIFT keypoints. Using a benchmark face gender classification dataset recently proposed in the literature, we obtain a state-of-the-art accuracy of 92.5%, thus validating our approach.
©2008 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.