Abstract

A new method of watch list screening based on the Bayesian framework, which can tell the probability confidence, is proposed in this paper. First, the AR database has been adopted as the training set to conduct the Principal Components Analysis (PCA). Then the whitening distance and its standard variance of each sample in the gallery set are calculated through the use of eigenvalues and eigenvectors of PCA. When a test face image is in, its confidence among the gallery samples can be gained. Thus, we can set a proper confidence threshold according to specific situations. After removing the results which are smaller than the confidence, results under different confidences can be reached. In this paper, two video databases with different confidence threshold have been adopted. The results gained in those two experiments are consistent with theoretical analysis, which demonstrates the effectiveness of the proposed method.

Paper Details