Jan. 23 2007 2:00PM

Jan. 23 2007 3:00PM

Many problems in computer vision can be (but rarely are) formulated as time-constrained optimization. We will discuss two example problems: (i) face detection and (ii) optimal randomized RANSAC algorithm for finding correspondence. We show how to derive quazi-optimal solution by applying Wald's theory of sequential decision-making.
In the face detection problem, we are interested in learning the fastest detector satisfying constraints on false positive and false negative rates. We solve the problem by combining Wald's sequential probability ratio test and AdaBoost. The solution can be viewed as a principled way to build a close-to-optimal "cascade of classifiers" of the Viola-Jones type. In the optimal randomized RANSAC, the goal is the fastest randomized strategy for hypothesis verification satisfying a constraint on the probability that the returned solution is correct. The optimal strategy is found again with the help of Wald's SPRT test.

Pravidelný seminář Rozhodování a řízení za neurčitosti. (Kontaktní osoba: Miroslav Kárný)