Research On Multi-feature Iris Recognition Algorithm Based On Random Forest

Biometrics technology mainly refers to a technology of identity authentication through human biometrics.By obtaining one or more of human biometrics,and converting them into digital information stored in the computer,the reliable matching algorithm is used to complete the process of verification and identification of personal identity.Biometric identification is more secure,and convenient than traditional identification methods.Biometric identification technology has the advantages of not easy to forget,good anti-counterfeiting performance,not easy to stolen,carry and can be used anytime and anywhere.The meaning of biometrics is very broad,which can be roughly divided into two categories:physical characteristics and behavioral characteristics.Physical features include fingerprints,veins,palm shape,retina,iris,body odor,face shape,and even blood vessels,DNA,bones,etc.Behavioral characteristics include:signature,voice,walking gait,etc.Iris is a special biological feature,which is a circular visible area between the pupil and the sclera.It is composed of a complex fibrous tissue that can expand and shrink with the change of the pupil diameter.It contains a lot of intersecting spots,filaments,coronae,stripes,recess and other details.After the iris is formed during fetal development,it remains unchanged throughout life.These characteristics determine the uniqueness of iris features,and also determine the uniqueness of identity recognition.Therefore,the uniqueness and stability of iris are unmatched by other biometric features,which makes iris become the mainstream product in the human biometric identification market.Iris recognition is to make use of the texture features of iris region in the human eye image to form a feature template and complete the recognition by comparing these feature parameters.The process of iris recognition mainly includes five parts:collection,location,segmentation,feature extraction and matching.Before positioning,due to the iris image has eyelashes,eyelids and other interference,all the image preprocessing,median filtering is commonly used in the stage.The positioning is mainly to identify the iris ring between the pupil and the sclera.The mainstream method to extract the iris ring is Hough circle detection and Canny edge detection by using gray value change.The segmentation process is to divide the extracted iris ring into a fixed size matrix array,and the normalization processing is convenient for subsequent feature extraction.Iris image feature extraction,mainly to extract the amplitude or phase information of the image,the most widely used is to extract the amplitude information as a feature.The matching process is mainly to compare the samples with all the samples in the database one by one,and the database sample with the highest similarity is the matching result.In order to extract features effectively and improve the recognition rate and stability of iris recognition system,an effective multi-feature iris recognition method based on random forest is proposed in this paper.Iris features were extracted by using Local Binary Pattern(LBP),S-transform and 2D-Gabor wavelet transform.The normalized iris features were extracted by LBP of different window units,2D-Gabor transform of different directions and S transform of different frequencies,and three groups with the highest recognition rate were selected as iris features.The training model uses the fusion of three feature templates to build multiple decision trees,and in the process of identification,the random forest method is adopted to randomly select attributes for matching.Euclidean distance is used as the identification criterion in the process of matching each sample with other samples.In the experiment,CASIA V3.0 database of Chinese Academy of Sciences and Iris_SDU database built by ourselves were used to recognize iris respectively.The experimental results show that the average recognition rate of the system can reach more than 99%in V3.0 database and more than 95%in IRIS_SDU database.So the randomness of random forest improves the recognition rate and generalization ability of the recognition system and makes the system more stable.