Anti-faces for detection

Abstract

This paper offers a novel detection method, which works well even in the case of a complicated image collection -for instance, a frontal face under a large class of linear transformations. It was also successfully applied to detect 3D objects under different views. Call the class of images, which should be detected, a multi-template. The detection problem is solved by sequentially applying very simple filters (or detectors), which are designed to yield small results on the multi-template (hence \anti-faces"), and large results on \random" natural images. This is achieved by making use of a simple probabilistic assumption on the distribution of natural images, which is borne out well in practice, and by using a simple implicit representation of the multi-template. Only images which passed the threshold test imposed by the first detector are examined by the second detector, etc. The detectors have the added bonus that they act independently, so that their false alarms are uncorrelated; this results in a percentage of false alarms which exponentially decreases in the number of detectors. This, in turn, leads to a very fast detection algorithm, usually requiring (1 + δ)N operations to classify an N-pixel image, where δ < 0:5. Also, the algorithm requires no training loop. The suggested algorithm's performance favorably compares to the well- known eigenface and support vector machine based algorithms, and it is substantially faster.