Zygomatic Smile Detection: The Semi-Supervised Haar Training of a Fast and Frugal System: A Gift to OpenCV Community
Daniel Devatman Hromada, Charles Tijus, Sébastien Poitrenaud, Jacqueline Nadel · 2010
Five different OpenCV-compatible XML haar cascades of zygomatic smile detectors as well as five SMILEsamples from which these detectors were derived had been trained and are presented hereby as a new open source SMILEsmileD package. Samples have been extended in an incremental learning fashion, exploiting previously trained detector in order to add and label new elements of positive example set. After coupling with already known face detector, overall AUC performance ranges between 77%-90.5% when tested on JAFFE dataset and <;1ms per frame speed of smile detection is achieved when tested on webcam-obtained videos. Observed results indicate that a semi-supervised incremental method which implements an existing haar-feature based classifier in order to to extend a sample from which a new classifier will be derived can be a method of 1) performance augmentation and 2) reduction of the amount of work dedicated to manual labeling of regions of interest in the positive example sample.