Iris Texture Features Analysis And Recognition Algorithm Evaluation

There has been a rapid increase in the need of accurate and reliable personal identification infrastructure in recent years. Biometrics has become an important technology for security. Iris is the colored part round the pupil of the eye, which is unique, stable, and inoffensive and can be collected easily. Iris recognition is an identification method based on texture features of the human eye iris to determine the identity, it is one of the most accurate biological recognition methods, and it has been applied in the security domains such as identity authentication. Compared with other biological specificity such as face and fingerprint, iris patterns are more stable and reliable. Furthermore, iris recognition system is non-invasive to the users. So the iris recognition technology has become the research focus in the current biological recognition region.Iris recognition system can be divided into four modules: image acquisition, preprocessing, feature extraction and classification module. In this paper, each module has been simply introduced, and then a simple evaluation was done in several ris recognition algorithms. The key issues in iris recognition system are the analysis of texture features and classification, which are also focused in our research. After an overview of previous research, the paper investigates the extraction algorithms for iris texture and classifier design. The main work as follows:1. An iris recognition algorithm based on lifting integer wavelet transformation is presented. Compared with the traditional wavelet transformation which was based on convolution, the algorinthm is simple in the operation and use the less memory, then the operation speed is improved and the transformation from integer to integer can be realized, which is beneficial to quantify the iris information.2. Each sub-band of wavelet decomposition contains plentiful information steming from different directions and frequencies of original image. According to the characteristics of the iris image, we regard low-frequency sub-band as a new iris image, and then use the Log-Gabor filter to extract features of new image. The hamming distance is employed to carry on iris recognition. The experiment results show the algorithm has the better effect than the algorithm only using the wavelet transform.3. Hamming distance is a simple, effective and popular classfier, but sometimes it dosen't get the optimal results. Support vector machines (SVM) are a new method of statistical classification, and the relative theory and application are rapidly developing. SVM can effectively deal with data sets of less samples, which has the better generalization ability. Therefore, SVM is employed as a classifier in this paper, the effectiveness of which has been shown in our experimental works.4. As the current evaluation methods for iris recognition are too unilateral, more comprehensive indices system is introduced in our research. Some typical algorithms of iris recognition have been evaluated by using these indices.