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 is also successfully applied to detect 3D objects under different views. Call the collection of images, which should be detected, a multitemplate. The detection problem is solved by sequentially applying very simple filters (or detectors), which are designed to yield small results on the multitemplate (hence, "antifaces"), 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. Only images which passed the threshold test imposed by the first detector are examined by the second detector, etc. The detectors are designed to act independently so that their false alarms are uncorrelated; this results in a false alarm rate which decreases exponentially in the number of de tectors. This, in turn, leads to a very fast detection algorithm. Typically, (1 + δ)N operations are required to classify an N-pixel image, where δ < 0.5. Also, the algorithm requires no training loop. The algorithm's performance compares favorably to the well-known eigenface and support vector machine based algorithms, but is substantially faster.
|Number of pages
|IEEE Transactions on Pattern Analysis and Machine Intelligence
|Published - Jul 2001
Bibliographical noteFunding Information:
This paper was substantially modified and extended following both reviews of an earlier submission to the ECCV 2000 conference and to IEEE Transactions on Pattern Analysis and Machine Intelligence. The authors are grateful to the reviewers for their insightful comments. The authors also thank Henry Rowley for providing the face images used for the experiments in Section 3.1. Part of this research was conducted while D. Keren and C. Gotsman were working for the Hewlett-Packard Company. This work was partially supported by the Israeli Ministry of Science under grant 1229.
- Distribution of natural images
- Image detection
ASJC Scopus subject areas
- Computer Vision and Pattern Recognition
- Computational Theory and Mathematics
- Artificial Intelligence
- Applied Mathematics