Quantitative Measurement of Face Detection Algorithm Performance

Measuring accuracy of a face detection algorithm hardly ever mentioned in detail and explicitly. The standard and terminology used in each experiment reported in each paper were vary and not standardized. In this paper, a method for calculating detection accuracy empirically is proposed. It consists of two algorithms that can be used for measuring accuracy of single face image and multiple faces image detection. Experiment has been conducted by using DeWa, a multiaspect-based face detection method, to measure four facial databases. The results are presented using standard terminologies for detection algorithm such as detection rate, false positive and true negative rates.