Research And Implementation Of Face Image Recognition With Kinect

Face recognition has been widely applied in human-computer interaction,security surveillance, access controlling, public security and so on, because of its non-contact, good stability and low cost of device in image acquisition. Traditionaltechnology of face recognition always uses2D images to recognize the human facialfeatures which is Susceptible to external interference. The Kinect body sensordeveloped by Microsoft could get the color data and depth data synchronously. Itprovides supports for2D and3D multi-modal face recognition.The main workcompleted in this article are as follow:2D face detection and recognition with Kinect is studied in this dissertation. Theacquired2D image is filtered by Gauss Pyramid and then the down-sampling isoperated. The feature of the face is described by Haar feature and integral image. Thestrong classifier is constructed by Adaboost algorithm and associated with each otheraccording to screening cascade. Then the face could be detected from the backgroundrapidly and accurately. In the process of face recognition, Local Binary Patternsoperator could describe the low resolution image well. By doing some experiments,comparing with the algorithms of principal component analysis and lineardiscriminant analysis, it proves that LBP can significantly improve the accuracy offace recognition.3D face detection and recognition with Kinect is continued in this dissertation,also. It uses the skeletal tracking technology to detect the human face. It couldsegment whole face areas from depth images with color information by regionalgrowth method. The acquiring depth faces should be filtered by Moving Least Squaresalgorithm. The last face recognition step is doing Iterative Closest Point algorithmand outputs results. Besides, threshold and point cloud number effect of facerecognition rate are discussed.Moreover, a decision fusion of2D face recognition results and3D facerecognition results is done by fuzzy integrals. The realization of multimodal facerecognition system software is based on VS2010.