A Method for Object Detection Based on Pixel Intensity Comparisons Organized in Decision Trees

The basic idea is described in the following paper:

N. Markuš, M. Frljak, I. S. Pandžić, J. Ahlberg, R. Forchheimer
A method for object detection based on pixel intensity comparisons, arXiv, May 2013

For people who wish to test the method themselves, the source code of the face detector is available on GitHub. A compiled executable file for windows-based machines can be downloaded here. As modern JavaScript engines became very fast recently, our face detector can run in web browsers: a self-contained HTML page (generated with emscripten).

Pixel intensity comparisons can easily be rotated for a given angle without the need for image resampling. This means that we can make fast rotation invariant object detectors. We demonstrate this by detecting rotated faces: demo applications.

A video demonstrating our results is available here (or on YouTube).