Research On Face Detection And Recognition Based On SVM

As an important biometrics technology,face detection and recognition has made great research results over decades of research.But due to the complex patterns of face detection and recognition and the vulnerability of detection images to external environmental factors,the recognition effects are still unsatisfactory,Therefore,the further improvement and perfection of face detection and recognition method will still have great application value.In order to improve the effect of face detection and recognition,this thesis focuses on the algorithm of face detection and recognition.It firstly gives a brief description of the research significance of face detection and recognition,and the research status in domestic and abroad.Then this thesis introduces the related technology used in the process of detection and recognition,and analyzes the problems existing in the current traditional algorithm.Finally,some improvement schemes are put forward for the algorithm in face face detection,feature extraction and face feature classification based on Support Vector Machine classifier to achieve better detection and recognition results.The specific research contents are as follows:In the face detection part,aiming at the problem that the Adaboost algorithm increases the error detection rate,improves the detection rate,a method of face detection which combines skin color with Adaboost is presented.Firstly,face color models are segmented,and candidate windows is set by skin color detection results so that the time of setting window size is shortened.Then Adaboost is used to get face location from skin color region.The results show that the combined method significantly reduces the error detection rate and improves the detection speed.In the feature extraction part,in order to solve the problem of information redundancy and dimension disaster,the Gabor algorithm is proposed by using the method of LBP extraction feature.At the same scale,the average values and amplitude in different directions are compared in turn to the face feature,and the face features in different directions are merged to reduce the feature dimension.The results show that the algorithm reduces the feature dimension and effectively improves the effect of feature extraction.In the face recognition part,aiming at the problem of linear inseparable in one-to-one SVM classification,the improved classification method of one-to-one voting structure SVM is proposed.When multiple maximum voting categories are obtained,the final classification result is determined by calculating the distance between the test sample and each of the maximum number of turnout categories.The results show that the algorithm solves the problem of linear inseparable and improves the recognition rate.The simulation experiment of the above algorithm is realized in MATLAB2012 B environment using ORL and combinatorial face database,The simulation results are compared and analyzed with detection rate,error detection rate,detection time and recognition rate as evaluation indexes,and the experiment shows that the algorithm is feasible and effective.