Research And Implementation Of Face Recognition Algorithm Based On OpenCV

Face recognition was widely used as biometric technology. It belongs to the field of patternrecognition of images, and then is a very high dimensional and typical big data volumeapplication case of pattern recognition. For a long time, the researchers on the study of facerecognition have never been interrupted. As smart surveillance, human computer interaction,security, and so on, face recognition has been widely used in many aspects. It becomes anindependent research subject.The paper introduces the theory of automatic face recognition technology. Then, expoundsthe real-time detection of the human face, face representation, feature extraction (featuredimension reduction), matching recognition, and so on. Finally, gives the algorithmimplementation. Specific research contents and innovative works are as follows:(1) Image acquisition process is restricted by all kinds of interference and randomnoise, the paper introduces several kinds of commonly used image preprocessing methods, suchas gray level transformation, histogram equalization, image filtering and so on, and apply themto real time face detection, feature extraction, etc.(2) Introduces the real-time face detection that based on AdaBoost algorithm. By using acascade classifier, proposes a kind of weighted cascade classifier face detection algorithm. Theexperiment shows that this algorithm has improved detection rate and without any increase of thetime complexity.(3) Based on PCA, and2D-PCA algorithm is analyzed. Based on global, ROI and not ROI,a kind of weighted chunking2D-PCA algorithm is putted forward. Through continuouscumulative values, find the optimal solution finally. Experiments show that the algorithm isimproved the determine accuracy of the human face, and the speed of predicate is not inferior tothe2D-PCA algorithm.(4) By the research of statistical learning theory, inspired by PCA algorithm of supportvector machine (SVM), a weighted2D-PCA support vector machine (SVM) algorithm isproposed. When the sample size is not consistent and big sample number is far more than thesmall sample size, the algorithm make small sample’s classification accuracy more betterimprovement.In the paper, all algorithms are implemented based on the Intel’s OpenCV open-sourcecomputer vision library, and VS2010integrated development environment.