|
Nowadays, the application of biometric technology becomes more widespread, has penetrated into all areas of life. Iris recognition is considered to be a higher degree of accuracy of biometric technology, has achieved great development. Iris recognition system comprising: iris acquisition, image processing, feature extraction and matching. Earlier work heavily concentrated in feature extraction and matching, however, much of the iris system performance depends on the quality of the iris image, therefore, paper looks at the iris image noise characteristics, based on different theories proposed two algorithms detect noise . Meanwhile, the space for scene classification pyramid core, applied to the iris recognition system in an effort to establish a new feature extraction and recognition model, and achieved good experimental results. This research work mainly in the following aspects: a study of a new method of removing eyelashes and eyelids, the traditional method is to eyelids and eyelashes as two different types of noise were removed. And we use are long eyelashes on the eyelid this fact, the eyelashes and eyelid connectivity for this model, relying on the principle of morphological region growing, remove the eyelids and eyelashes, not only the use of gray-scale information, and gradient information to the removal of noise limitations. Two iris recognition system as a non-invasive system of high, often due to the introduction of a variety of user gestures and different noise. To expect various forms of noise is almost an impossible thing, so it is difficult to find a highly versatile method to detect various types of noise, based on this, we study a novel machine learning algorithm, the method FJ-GMM using algorithms to estimate the distribution of noise and the iris, and then using the corresponding Bayesian separator, the noise classification. This method can maximize the use of a priori information about the user, enables iris recognition system used in a variety of complex environments, but also because accurate denoising improve system performance. 3 presents a new iris feature extraction and matching model. This method features and pyramid match the BOW nuclear combine spatial pyramid model was established. Methods edge of the iris image as the characteristic points, while the maximum of each edge point using the spatial location information. The normalized iris image maps eventually become a multi-scale histograms. The method is based on the core recognition method, combining kernel-based classifiers, such as SVM, etc., you can get good recognition results. Compared to other methods, but the method of calculation is greatly reduced and it may well be applied to real-time iris recognition systems. Algorithm used in the pictures are developed iris acquisition device collection to get. Meanwhile all the algorithms are run on CASIA standard library and compared.
|