Research On Glass Problem In Near Infrared Face Recognition

In recent years, biometric identification technology is the hot security means ofthe large scale security projects. Face recognition technology is one of the mostwidely used and effective ways in the security. Near infrared image is adopted byface recognition technology with its subjectivity. However, there are limits to thenear infrared face recognition in the applications. Glasses recognition is one of thelimits which affect the accuracy and speed of near infrared face recognition.In this paper, the influence of glasses problems on near infrared face recognitionis analyzed and the solution and corresponding experiment is presented in theenvironment of near infrared face recognition.Firstly, the characteristic of biometrics identification technology and nearinfrared image is introduced in this paper; Secondly, the hot pots of face recognitionresearch are presented; Thirdly, design two kinds of software, which preprocess theimage before the experiment, including eyes location and face normalization.In the part of feature boosting, AdaBoost is selected. For shortening the trainingtime, based on the characteristics, an improved Cascade structure is advanced,which discards both negative and positive samples. The new Cascade structureincreases the weighted value of critical samples. So AdaBoost focuses more on thesamples to be solved. And more than that, this structure shortens the time ofcalculating the error of weak classifier on samples. For these two reasons, thetraining time is shortened.In the part of feature usage, aimed at the disadvantage of Haar features, Gaborreal-part features and Gabor module features are added alternatively. Theexperiment result shows that such feature option not only increases the detectionrate of glasses in near infrared images, but also avoids the repetition of samefunctional features, which reduces the training time in another way.The highlight light problem in locating eyes is analyzed and concluded. Theproblem is classified into two situations, which are semi-blocking and completelyblocking. Based on the result of glassed detection in near infrared face images, anew thinking and solution is brought up, which can solve the highlight problem inlocating eyes with no need of taking off the glasses. The solution advanced theaccuracy rate of near infrared face recognition.