Study On Processing And Matching Of Fingerprint Images

The popularity of internet has boosted the demand of security based on individual identity. For its portability, uniqueness and high veracity, fingerprint is currently one of the most widely used biometric technologies. Although there has been published many researches about fingerpint recognition and many devises based on fingerprint have been designed for business application, fingerprint recognition is still a challenging problem due to the complex conditions existing in fingerprint image and limitations of real time application.This dissertation takes plenty of research materials about fingerprint recognition both at home and abroad for reference and analysis. And technologies like image processing, pattern recognition and machining learning are also employed for this research. In-depth studies are made in areas of fingerprint image segmentation, fingerprint image enhancement, fingerprint core point extraction and fingerprint matching. Some useful results are yielded in both fields of theory and practice. The contributions of this dissertation are summarized as follows.(1) Fingerprint image segmention is discussed. Features of fingerprint image are analyzed in three different aspects. According to the mean and variance of pixel gray value in local area, clarity of fingerprint ridges and symmetry of local ridges, intra-consistency, extra-consistency and local ridge conformity have been defined. Foreground and background of fingerprint image is differentiated by utilizing multiply features. The experimental results show that the proposed method can segment fingerprint image in an effective way, especially for the regions between foreground and background. Moreover the computation cost is relatively small.(2) Fingerprint image enhancement is analyzed. An improved Gabor filter based method for fingerprint image enhancement is proposed. Traditional Gabor filter based method uses a fixed filtering region, while a more reasonable and effective region selecting strategy is devised on the basis of frequency and orientation of local ridge in this dissertation. A fast implementation of Gabor filter is also employed for the symmetry of Gabor filter. The experimental results show that the proposed method is effective and the computation cost is reduced. (3) Method for fingerprint core point detection is investigated. A fingerprint core point detecting algorithm by using Supported Vector Machine (SVM) and tangent complex filter is proposed. In terms of ridge distribution in core point region and non-core point region, feature vector is constructed. Then SVM is used to form the classification model, and the candidate region in which core point exists can be predicted. Since the ridge around core point is homo-symmetry, a complex filter is designed on the basis of tangent function. By comparing traditional complex filter, tangent complex filter can depict ridges around core point more suitable. Hence fingerprint core point is localized with high resolution. The experimental results show that the proposed method can compute the position and orientation of fingerprint core point precisely. Moreover, the computation cost is also reduced.(4) With the existence of distortion in fingerprint images, the position of ridges can be changed to some extent. This dissertation designs a fingerprint matching method based on excluding elastic distortion. First adjacent feature union (AFU) is defined according to neighbor minutiae. Then reference minutiae pair is obtained by comparing AFUs between template and query fingerprints. Reference minutiae pair is used for fingerprint alignment. Ridge frequency and orientation between reference minutiae and query minutiae are used to construct the distortion model. The position and direction of query minutiae are adjusted according to corresponding distortion model. Then matching producure is carried out by adopting minutiae after adjustment. Experimental results show that our proposed method can reduce the effect of distortion in fingerprint image, and has high robustness.(5) Traditional fingerprint matching methods based on single reference minutiae can be affected by partly obtained fingerprint images and inevitable noises. This dissertation proposes a fingerprint matching algorithm by multi-reference points. First local feature unit(LFU) is constructed by use of minutia and its neighbor ridges. Potential matched minutiae set (PMS) is obtained by comparing LFUs of template and query fingerprints. True matched minutiae set (TMS) is obtained by purifying PMS. Minutiae in TMS are used as multi-reference points. The remaining minutiae are compared according to the distance from minutiae to TMS. The experimental results show the multi-reference points fingerprint matching algorithm is more effective and efficient than the single reference point fingerprint matching algorithm.