Viola-Jones Algorithm

The Viola-Jones algorithm is an object detection framework, developed originally for detecting human faces in real time, that combines simple rectangular brightness-comparison features with a boosted classifier arranged as a cascade. It scans an image with sliding windows and evaluates each one against a series of Haar-like features, patterns such as an eye region reading darker than the cheek below it, computing every feature almost instantly with the help of a precomputed running-sum image called an integral image; a modified version of the AdaBoost algorithm selects and weights the small number of features that matter most, combining them into a single strong classifier, which is then organized as a cascade of increasingly strict stages so that a window can be rejected as a non-face after only the first few, cheapest tests, leaving the expensive later stages to run only on the promising remainder. Paul Viola and Michael Jones published the method in 2001, and its combination of accuracy and real time speed made it the standard technique behind consumer face detection, including in early digital cameras, for roughly a decade afterward.