Research On Applying Fingerprint Verification To Access Control System

As an important branch of biometric verification, fingerprint verification (FV) hasadvantages of mature technology, high verification rate and reliability. With the rapiddevelopment of fingerprint sensor, computer and image processing, FV has been developedgreatly. The application domains of FV have expanded from criminal investigation toelectronic certificate, attendance machine, and access control system (ACS). The researchemphases of ACS are promoting algorithms performance and making hardware devicessmaller. It’s important to develop high-performance algorithms which can run on ARM orDSP. The main research contents of the paper are fingerprint acquisition, normalization,segmentation, feature extraction, and match.In fingerprint acquisition section, STM32F429and FPC1011F are chose as themicro-controller and image acquisition device separately. The new image acquisition systemhas higher cost performance than the system based on DSP.In preprocessing section, promoted normalization and segmentation methods are adoptedand proposed separately. When normalizing, divide image into sub-blocks, and use the meanand variance of block to adjust the desired mean and variance. Block normalization has betterperformance than global normalization. When segmenting, firstly, calculate gray and gradientthresholds with maximum between-cluster variance and extract region of interest (ROI)preliminarily. Then, use opening operation and annular mask to post process ROI. Theproposed segmentation algorithm has strong robustness.In feature extraction section, use texture feature from minutiae neighborhood (TFMN)and linear feature to supplement minutiae feature, and put forward new linear featureextraction method. To extract rotation-translation-invariant TFMN, circular complex filterneed to be constructed. To extract linear feature, track a base segment firstly. Then, grow thesegment along its direction with special conditions. The TFMN and linear feature are usefulfor ACS, because they could be extracted and matched easily, and have less memory desire.In match section, a hierarchical match method is proposed. Firstly, screen out minutiaethat maybe matched with TFMN. Then precisely match minutiae feature that screened out atfirst step. If minutiae match failed, use the linear feature to match. The TFMN match canreduce align time obviously, and liner feature match can also reduce false reject rate whenmatch fingerprints which are lack of minutiae.