Rapid face detection using a multi-mode cascade and Separate Haar Feature

In this paper, firstly, we describe a new feature for fast and accurate face detection. The feature is called Separate Haar Feature. Secondly, we describe a multi-mode detection algorithm to improve the detection rate. There are three key contributions. The first is "Separate Haar Feature", which adds a don't-care area between the rectangles of Haar Feature. We can get some more efficient features by this algorithm. The second is the algorithm for selecting the best width for this don't-care area and the False Alarm Rate for each stage in "Learning". This algorithm for width selecting is proposed to reduce the total number of learning features for reducing the memory used. And we use a smaller false alarm rate for each stage and less number of stages training the detector to improve the hit rate within the same detection time consuming. Finally, we proposed a multi-mode detection algorithm in cascade detection process to improve the detection rate.