Products >> Face Detection
Low false detection, high accuracy
Can simultaneously detect multiple faces
Can detect different color face
Can detect faces in a complex background
Written in C, can easily be ported
Based on MB-LBP(multi block local binary pattern) features lookup table type weak classifiers Real AdaBoost face detection algorithm. LBP (Local Binary Pattern) features proposed by the Ojala in 1994, and applied to the texture classification problem. MB-LBP feature is an extension of LBP, uses image blocks instead of the original LBP features which a single pixel as the basic unit. MB-LBP can reduce the image noise when calculate LBP features, if adopt integral image technique, it is possible to be obtained MBLBP features in constant computation time.
AdaBoost is a boosting learning methods, AdaBoost training process using the threshold as a feature of weak classifiers output, this weak classifiers has limited ability to divide sample space. Based on Real AdaBoost algorithm, Wu proposed a lookup table type weak classifiers continuous AdaBoost face detection algorithm, to get a good face detection results.
Research on face detection algorithm uses publicly UMass Face Detection Data Set and Benchmark (FDDB, http://vis-www.cs.umass.edu/fddb/) face detection database. MB-LBP lookup table type weak classifiers Real AdaBoost face detection algorithm and other published methods were compared, the results shown in figure, it can be seen from the figure, MB-LBP lookup table type weak classifiers Real AdaBoost face detection algorithm exceed other methods.
Comparison with the effect of the OpenCV face detection, as shown below. Left is MB-LBP detection result, right is OpenCV detection result. MB-LBP face detection has less error detection, high accuracy.