Study On Fingerprint Enhancement And Classification Method

Biometric recognition refers to the use of distinctive anatomical and behavioral characteristics,called biometric characteristics for automatically recognizing individuals.Compared with documents,passwords,command and so on,biometric characteristics are not easy to lose or forget,so they have better recognition ability and reliability.Automatic fingerprint identification system has many advantages,such as small volume,simple operation,low cost and high reliability.It is becoming more and more popular,and has become one of the most important biometrics recognition technologies.As important parts of automatic fingerprint recognition,fingerprint enhancement and fingerprint classification play important roles in the process of recognition.Fingerprint enhancement aims to improve the clarity of fingerprint ridge structures and to ensure the accuracy of the feature extraction in low quality region of fingerprint.Fingerprint classification aims to reducing the number of comparisons that are required to be performed in a large fingerprint database by classifying fingerprints into predefined classes,which is a significant technique of the fingerprint identification system.This paper mainly studies fingerprint enhancement and fingerprint classification of automatic fingerprint recognition.The main research work and contributions are as follows.1)In this paper,a curved-regions-based fingerprint enhancement algorithm in Fourier domain is proposed.The previous works enhanee fingerprint images in Fourier domain based on block.But there.are some shortcomings in these works,they are not robust enough to non-smooth areas,and the size of the window is not easy to determine.Mainly include following improvements:curved-region transform,which finds curved region and map to 2-D array,is used in Fourier domain,and filters are designed based on the frequency images of curved regions,curved-region transform locate;fingerprint image is partitioned into smooth areas or non-smooth areas,and only construct curved regions in non-smooth areas in order to improve the execution efficiency;composite window is introduced to resolve the choice of window size.Experiments show that:the improved method strengthens the robustness due to curved region transform and composite window,it has better enhancement effect than the algorithms based on block,especially in non-smooth areas.2)A fingerprint classification approach based on the DCT coefficients of orientation field is proposed.The proposed approach mainly includes the following steps:calculate the sine and cosine components of the orientation field and perform two-dimensional discrete cosine transformation;the low-frequency parts of the matrix which is gotten by the discrete cosine transformation are extracted as feature vector to describe approximate orientation pattern of whole fingerprint;fingerprint is classified through SVM.This algorithm mainly includes following advantages:the DCT coefficients of orientation field contain enough information to achieve classification,and the feature vector is more reliable;the length of the DCT feature vector is smaller,and the classification speed is faster;the algorithm has considerable classification effect on fingerprint images with different sizes,this algorithm does not need to locate the reference point and has higher reliability.Experiments show that the proposed method achieves satisfactory classification accuracy.