Design And Implementation Of Face Recognition System Based On Eigen Value

With new breakthroughs were made in Facial-Recognition Technology,its application value and commercial potential has gradually recognized by the whole society,and in urgent of automatic recognition system which charactered as deploy easily and perform stable.In recent decades,a large number of face recognition algorithm emerged— light pretreatment to eliminate the influence of light,feature extraction methods in face feature extraction,as well as many kinds of classifier and distance calculation method.Among those methods,combinations of histogram equalization processing?linear subspace feature extraction method and K-NearestNeighbor classifier,can keep a high identify accuracy on the premise of recognition fast,and became a very popular method.On the basis of predecessors' research,this paper did the following works:(1)Analyzed the business potential and development trend of face recognition technology,summarizes the research in recent years such as light pretreatment,face detection and recognition methods,briefly analyzes the advantages and disadvantages of them.And listed face databases in different standard.(2)Based on "Haar features and Adaboost classifier" detection method,realized a real-time face detection module.In the front of the face detection step,light pretreatment and noise filter were set to optimize the input images,ruled out some parts of illumination and noise influence during the face detection step.(3)Respectively explored the PCA and 2DPCA feature extraction method,and modeled the two algorithms in Matlab,and tested on the ORL and FERET face databases,also conducted a lot of experiments in the test,constantly adjust algorithm method through the experiment result.And got a ideal algorithm model satisfy recognition speed and accuracy two factors finally.(4)To overcome the same count votes defects in the vote link of the original K-Nearest-Neighbor classifier,improved K-Nearest-Neighbor classifier was put in the paper,and analyzed the face recognition system performance under changing conditions such as training images number ? feature dimension and sample capacity,experience results have proved that the improved K-Nearest-Neighbor classifier can make the system to obtain better identification effect.(5)On the basis of the research and modeling algorithm implemented a real-time face detection and recognition system,the system has a certain robustness during the expression verify?illumination?or position change.