Figure 14

Comparison of different strategies to estimate posterior probability in a boosted McCascade. They are for different rates of missing weak classifiers. In (a), 50% of weak classifiers are missing, while in (b), 60% are missing. Each McCascade can be compared with the naive approach presented in Section 3.1 where the initial boosted cascade is used with available weak classifiers (noted CascadeA). In each curve, we also plot the performance of the boosted cascade when all weak classifiers are known (noted CascadeF) to show the effect of missing weak classifiers on initial performance (best view in color).