The History of AI

SIFT and the Viola–Jones face detector

Era 3 · From the expert boom to the statistical turn (1980–2011) › Robotics and perception

Lineage
Acting in the world
Kind
Publication

Who

University of British Columbia, Mitsubishi Electric Research Laboratories

What happened

In 1999 Canada's David Lowe presents SIFT, a method for finding keypoints in an image that can be recognized even when the scale or rotation changes. In 2001 Paul Viola and Michael Jones present the first face detector that works in real time.

Why it mattered

Hand-designed features and statistical learning bring computer vision to cameras and apps. These are the methods that dominate the field until convolutional networks overtake them in 2012.

The full story

Recognizing an object in a photo is hard because the same object shows up larger or smaller, rotated, under different light, or partly hidden. In the 1990s the most promising route was to hand-design descriptors, numerical summaries of small patches of the image that would not change with those variations.

Lowe, at the University of British Columbia, found a very effective one. SIFT locates distinctive points at several scales and describes each one with a histogram of the directions of the edges around it. Two photos of the same object share many of those points even when they were taken from different places. The method was used to recognize objects, stitch photos into panoramas, and help robots find their bearings.

Viola and Jones tackled a different problem: finding faces in an image fast enough for video. They used very simple features, differences in brightness between rectangles, and let AdaBoost pick out the few useful ones from tens of thousands. A cascade of classifiers quickly discarded the regions that were not faces. The system processed 15 frames per second on an ordinary computer, and face detection made its way into consumer cameras.

In 2005 Navneet Dalal and Bill Triggs added histograms of oriented gradients for detecting people. These are the methods AlexNet surpassed in 2012.

Builds on

  • AdaBoost (1995)

    Documented direct influence

    The Viola-Jones detector uses AdaBoost to choose its features, as the paper's title says

Sources

How to cite

“SIFT and the Viola–Jones face detector.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/sift-viola-jones

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline