The Research Of Disjunctive Normal Random Forest Algorithm On Face Recognition

As one of the most challenging research subjects,face recognition has been greatly concerned in recent years.Face recognition is a kind of technology,which is about extracting the facial features through the computer and to authenticate identities according to these features.It has been a long time since people began to study face recognition.But how to apply these methods to meet identifying requirements on different conditions is still the focus of the recent study.This thesis studied and summarized the existing face recognition methods,and introduced the concept of disjunctive normal form.A kind of face recognition method which is based on random forest disjunctive normal form is proposed.This thesis' s main research work and innovations are as follows:Firstly,face recognition methods local and abroad were summarized and analyzed.Several popular methods of feature extraction and classifier algorithm were described in detail.Secondly,the model which is based on disjunctive normal form was put forward to effectively integrate random forest algorithm of weak classifier and improve the classification ability of the model.In this paper,the thought of disjunctive normal form was used in random forest algorithm.After integrating the weak classifiers in random forest,and then building the global objective function using the thought of back-propagating,all weak classifiers in the model can be studied under a unified framework.The experimental results show that using the thought of disjunctive normal form in random forest algorithm can help to effectively improve the classification performance of the models,reduce the generalization error of the model.Then,the strategy of comparative experiments was adopted to test training time and error rates.In addition,in actual experiment,there may be the situation of artificially misstating wrong samples or labeling the edge samples by mistake.So a certain number of fault samples were added to verify the model's ability to tolerate wrong samples and to saw if there was a big change on classification accuracy after joining error samples.The experimental results show that comparing with the random forest algorithm,disjunctive normal form of random forests algorithm had a stronger ability of tolerance error samplesFinally,the method of directions two-dimensional principal component analysis was adopted to extract facial features of face images.And face images were recognized with disjunctive normal form of random forest model.Contrast experiment was adopted to verify the advantages of disjunctive normal form random forest in face recognition,through the comparison of experimental results and random forest model.The proposed disjunctive normal form random forest is feasible in face recognition through the results.The proposed model is better than random forest model on classification performance.So comparing to the random forest model,it has a better behavior on recognition rate.