Research On Extraction And Recognition Of The Gabor Iris Features Based On SVR

The use of biometrics for identity identification is widespread concern in the field ofinformation security. In various biological characteristics, the iris for its uniqueness,universality, invariance, etc. become an effective identification tool.In this paper, we focus on the part of extraction and recognition in the process of irisrecognition in the background of iris recognition industry. For the extraction methods, we useGabor filter on iris images from CASIA V1, CASIA V4-Interval, JLU-IRIS2iris imagesdatabase to get the transform domain features, and proposed the way to encoding multiple irisfeatures in transform domain so that iris texture information can be efficiently organized as afeature vector to be matched. For the identification methods, we integrate the results ofmatched features into a vector as the input of SVR, and use the training result to identifysamples.The main works of this paper are as follows:(1) Segmented the ROI of iris images after pretreatment to obtain images eliminatedinterference that could reduce the impact of feature extraction.(2) When the iris images features are extracting, we get the information of amplitudeand phase through filtering images by Gabor filters. For this information, we use a new blockencoding method to represent them: blocked matrix of the amplitude and use the mean of eachsub-block of the matrix as the block coding; blocked the real part and the imaginary part offiltering results and code0or1for each sub-block. Then, in the experiments with featurediscrimination as an evaluation criterion compared local features using the proposed encodingmethod and global features which are not used to illustrate the effectiveness of the method inthe experiments. Meanwhile many of these different characteristics in blocks were compared,and found the best way to block the corresponding iris library.(3) Proposed a new method for discrimination by fusion results of matching multiplefeatures. Matched the various features and the results were divided into two subsets which areintra-class and inter-class, and combined into a vector for SVR. Then, set the correspondingregression values0and1. Such fusion model obtained through training to treat samples toidentify discrimination. In the experiments, the recognition result of fusion has been raised incontrast to a single feature and proved that the use of multi-features fusion algorithm is effective.In summary, this paper presents a new efficient Gabor iris feature encoding and applied itin multiple iris image databases to obtain the corresponding best way to block. Proposed aSVR fusion model of results of matching multiple features, and demonstrated theeffectiveness of the algorithm through experiments.