**Abstract**

Many classes of images have sparse structuring of statistical dependency. Each variable has strong statistical dependency with a small number of other variables and negligible dependency with the remaining ones. Sparse structure simplifies the task of recognizing various objects. In particular, a semi-naive Bayes classifier compactly represents sparseness. A semi-naive Bayes classifier decomposes the input variables into subsets and represents statistical dependency within each subset, while treating the subsets as statistically independent. This talk describes an automatic method for constructing a semi-naive Bayes classifier for object detection. This method generates a pool of candidate subsets where each subset captures a significant statistical dependency. The method then trains a log-likelihood function over each such subset. A group of these log-likelihood functions are selected to form the final classifier based on cross-validation performance. This approach achieves reliable and efficient detection for several objects including faces, eyes, ears, telephones, push-carts, and door-handles. |

Charles Rosenberg Last modified: Mon Dec 9 10:46:20 EST 2002