ViolaJones

Viola–Jones is a framework for fast object detection introduced in 2001 by Paul Viola and Michael Jones. It is best known for real-time face detection in grayscale images and video, achieving high frame rates on conventional CPUs by using a cascade of increasingly complex classifiers.

Core components include Haar-like features, an integral image representation for rapid computation, AdaBoost learning to select a compact set of features, and a multi-stage cascade that quickly rejects non-face windows. During detection, a sliding window scans the image at multiple scales; each window passes through successive cascade stages, with many windows discarded early to maintain speed.

Training involves assembling many positive face samples and negative images, then boosting to select discriminative features and weights. Features are computed from rectangular regions, such as edge, line, and four-rectangle patterns; the integral image allows calculating feature sums in constant time regardless of rectangle size.

Strengths of the Viola–Jones method include real-time performance on modest hardware and good accuracy on well-lit, frontal faces. Limitations include sensitivity to lighting changes and pose, occlusions, and variability in facial appearance; performance depends on the training data, and it tends to underperform compared with modern deep learning detectors on challenging data.

Impact and legacy: the approach popularized cascade classifiers and Haar-like features, influenced many early open-source computer vision implementations (for example OpenCV) and shaped the development of rapid detection methods, even as newer deep learning techniques have surpassed it in many applications.