Towards a cleaner method for multi-view face detection

Object detection is a very active research area in the eld of computer vision. A number of machine learning techniques have been applied to the problem with varying degrees of succes. The early implementations used arti cial neural networks, while recently most apporaches have explored boosting simple features. Viola and Jones were extremely successful in this approach, their paper is one of the most cited articles on the subject.[5] Viola and Jones introduced several innovations. First they used the integral image to compute simple Haar features that are extremely easy to compute, yet can encode relevant information. A pixel in the integral image is de ned as the sum of all pixels above and to the left of the given pixel in the original image: