The Study Of Dual-Mode Fingerprint Recognition Algorithm Based On Image Dimensionality Reduction

The fingerprint of human has its own property like unique, never change and can't be separated from the main body, and can satisfy the requisition of identification, so, researchers pay close attention to the identification technology based on fingerprint feature. Fingerprint recognition system is applied in kinds of secure areas, and it has been the one of most popular biological feature recognition technologies. But during the not good enough situation, the performance of fingerprint recognition system has many shortages which brings lots of difficulties, for example, the fingerprint which is too dry or too wet will lead to disconnection or attachment in the ridge lines, the difference of strength or direction during the finger pressing will lead to distortion. So, from the pre-processing of low quality fingerprint to the real-time of massive data system, the fingerprint recognition system has many key technologies to be solved.Based on the analysis of currently research result, this thesis will propose some views and improvement to the pre-processing, dimensionality reduction and feature matching of fingerprint image.The pre-processing of the fingerprint system can be divided into several main modules, as segmentation, enhancement, binarization, thinning and quality estimation. This thesis introduces the key parts of the pre-processing system in detail, makes an experiment or improvement on these algorithms, and gives the result finally.The dimensionality reduction of image map the source data from high-dimensional space to low-dimensional space, through maintaining the most important or individual variation property, making the data after reduce dimensionality can be as much as possible to keep the features of high-dimensional vector or the property of certain aspect. This thesis analyzes and studies on the PCA and 2DPCA algorithm of the dimensionality reduction and recognition of image, and adds the information of sample classification into the algorithm.The fingerprint matching algorithm has two patterns, which is based on local information and the other is based on global information. First, this thesis introduces and analyzes the traditional point pattern matching algorithm based on minutiae information; second, studies on the method of image dimensionality reduction and its application in fingerprint recognition; at last, proposes a dual-mode fingerprint recognition algorithm based on image dimensionality reduction. The algorithm has the advantages of both two matching patterns that can accelerate the processing of fingerprint matching and increase the accurate rate to a certain extent.The main points and innovation are as follows:(1)The algorithm process and realize steps of a complete dual-mode fingerprint recognition system are introduced and the relevant analysis and improvement based on the exist algorithms is proposed.(2)This thesis does further study on the pre-processing of fingerprint, and improves some algorithms as follows:1, A new algorithm to calculate the local threshold value is proposed to the segmentation of fingerprint;2, A better algorithm with block equalization is proposed to the equalization of fingerprint;3, An effective fast thinning algorithm is proposed to the thinning of fingerprint;4, An algorithm which adopts the block orientation field point sets is proposed to the core point location of fingerprint.(3)The dimensionality reduction and the feature extraction of image are studied, the 2DPCA algorithm is improved, and a 2DPCA algorithm based on the sample classification information is designed.(4) According to the characteristic of the fingerprint image, the improved 2DPCA algorithm is used in the fingerprint recognition, and a fingerprint recognition algorithm based on improved 2DPCA algorithm is proposed.(5)The point pattern matching algorithm based on minutiae information and the improved 2DPCA algorithm based on global information of fingerprint image are combined, and a dual-mode fingerprint recognition algorithm based on the dimensionality reduction of image is designed.