Research On Fingerprint Identification System

Fingerprints have the characters of uniqueness and invariance, which provides an ideal resolution for the automatic identification system. Fingerprint identification has become one of the most popular, convenient, reliable personal identity authentication technologies. Howerver, fingerprint indetification technology is still a very difficult task for some factors, such as noise, elasticity.The main steps of fingerprint identification system are image preprocessing, feature extraction and matching. This dissertation focuses on the following contents and makes some improvements:Firstly, there are five steps in the image preprocessing procedure, which are quality estimation, image segment, image enhancement, binarization and thinning. In this dissertation, an improved algorithm for the fingerprints enhancement based on Gabor filter is presented. We propose to use non-coherence factor to find the blocks whose ridge orientation are highly non-coherence, and apply 8-orientational Gabor filters to these blocks. Through orthogonal decomposition, we convert 2D Gabor filter into two 1D Gabor filters. This approach reduces the cost of time and improves the quality of fingerprint image.Secondly, there are three parts in the minutiae extraction, which are minutiae points detection, singular points detection and false minutiae elimination. 8-neighborhood feature points extraction algorithm is used to detect minutiae points. Singular points extraction based on the analysis orientation field is improved. This method can get singular points'position exactly.Thirdly, on the basis of the analysis on traditional indetification, a fingerprint minutiae first matching method based on the Delaunay triangulation is proposed. The next matching step is based on radial basis functions which can affine local deformation. Our method can reduce the cost of time and improve the precise of matching.Finally, we adopt matlab 7.0 to simulate the algorithm, and adopt FVC2000, FVC2004 as the database. The result shows these algorithms can work effectively.