Research Of Recognition On Feature Level Fusion Of Face And Iris

Biometric identification technology is a solution method of identification system relies on the body's physical characteristics including face, iris, fingerprint, hand type, palm-prints, voice prints, and DNA and other human characteristics. Through the sampling of biological characteristics, biometric identification system draws the only feature and converts it into digital code, and then combines these characteristics as the code template. When people interact with the identification system for authentication, the identification system obtain the characteristics and compare it with the feature template in the data base so as to justify the matching degree and determine whether to accept or the person or not.The identification technology based on one biometric is called single-mode biometric identification technology. Single-mode biometric feature identification generally has the following limitations such as: noise of samples, intra-class variations, inter-class similarities, non university, and lower spoof attack. So far, the identification system based on single-mode biometric is unable to fully meet the requirements of practical applications. The proposed multi-modal biometric identification technology could overcome the shortcomings of single-mode biometrics and makes an important research direction for the biometric identification technology. Multi-modal biometric identification technology is an identification technology using a variety of biological characteristics. The learning object of multi-modal biometric identification can be the same biological characteristics and also a variety of different biological characteristics. The study of fusion of a variety of biological characteristics o is truly multi-modal biometric recognition technology.The fusion level of multi-modal biometric feature according to typical biometric identification systems could generally be divided into pixel level, feature level, matching level and decision-making level, of which the amount of information is decreasing. From the realization difficulty of view, decision-making level is the simplest part that requires only a simple logic to determine. In specific applications, each fusion method has its own advantages and disadvantages. Although the feature level fusion method has large increase in efficiency, but it is not easy in the integration of ready-made single-mode biometric authentication system. Matching level fusion achieves the integration of information with little difficulties, it has good application value. Although the decision-making level holds a low degree of information integration, it has relatively little difficulties of integration so that could be used in a number of systems.According to requirements of multi-modal biometric fusion application, it needs to make a choice of the fusion objects and the fusion strategies. In a short word, three following issues should be solved: what is the theoretical basis that the multi-biometric fusion could improve the system performance, how to select the multi-biometric and the fusion strategies according to the actual application requirements, how is the effect of the after multi-biometric and the fusion strategy is given. The solution of above three problems will settle a clear theoretical basis of multi-biometric feature fusion strategy. It has practical significance of the in the construction and application of multi-biometric feature fusion system.This paper researched and improved feature level fusion algorithms based on face and iris through studying some existing face feature extraction algorithms, iris feature extraction algorithms, feature level fusion strategy and pattern recognition, image processing, multi statistical theories, and referencing a lot of literatures inside and outside country. Experiments showed that good effects had been achieved.This paper first studies the research status of multi-modal biometric fusion, and gives the advantages and disadvantages of different fusion level from the performance of the amount of information, the fault-tolerant capabilities, the implementation difficulty and the improved efficiency, then analyse the fusion strategy in different levels. After the comparison of every single biometric feature face and iris is final select as the input of feature level fusion. Through the study and comparison different feature extraction method of face and iris, the paper propose the multi-biometric feature level fusion model based on human face and iris.After the establishment of fusion model, the paper takes the classical canonical correlation analysis method in multivariate statistical analysis as the fusion strategy to realize the fusion of two different features. The idea of canonical correlation analysis derives from another classic method of data analysis in multivariate statistical analysis-principal component analysis. The paper promotes the idea of extracting the principal component which is unrelated from one set of variables to two sets.With canonical correlation analysis in order to function as the maximum correlation coefficient of the objective function, find the two sets of variables related to the largest typical projection axis, projected onto the projection axis find the typical integration of relevant variables is a new feature, as the next stage of classifier input. With the maximum correlation coefficient function as the objective function, the canonical correlation analysis method finds the typical projection axis of two setFrom the maximum correlation coefficient function defined in traditional canonical correlation analysis, we know that in the solution process is often encountered in the covariance matrix singular and higher-dimensional small sample situation, so that may restrict the using of canonical correlation analysis method. In the last chapter, the paper verifies the using of canonical correlation analysis method under the ideal conditions of the non-singular matrix. To solve the problem effectively, chapter 4 propose an integration of strategy based on the improvement of canonical correlation analysis. Under the thought of striking the canonical projection matrix, the method takes two sets of feature as the study object, and take the canonical correlation projection matrix after as the new fusion feature output to the next stage classifier. The method obtains more information of the feature matrix, and reduced the dimensions of the covariance matrix so that the computational complexity has dramatically decreased, and the efficiency of the final identification has increased.In conclusion, this paper studies the multi-modal biometric fusion levels and fusion method for human face and the iris. After the in-depth analysis and comparison of several commonly used feature extraction methods, the paper verify the multi-modal feature fusion model based on the canonical correlation analysis fusion method has a better recognition results. Based on the traditional canonical correlation analysis method, the paper later proposes a improved feature fusion algorithm of high efficiency. With the continuous development of science and technology, multi-modal biometric recognition technology has also gradually improved the author hope that this work carried in this paper could make some contribution to the development of multi-modal biometric recognition technology.