Research On Algorithms Of Preprocessing And Minutiae Extraction Of Fingerprint Image

The technology of fingerprint identification, widely applied in a variety of situations of personal identity verification and recognition, has made a great progress. However, there are still challenging problems, such as enhancement of poor quality images, extraction of fingerprint features, which remain not well-solved. The process of fingerprint identification can be divided into three steps: preprocessing of fingerprint images, feature extraction and feature matching. The thesis focused on algorithms in first two steps, including direction calculation, binarization, lines healing and feature extraction. A block direction calculation method was proposed, which used the gradient smoothness based on Sobel operator. This algorithm enhanced the sensitivity to the areas in which the curvature of the lines in fingerprint image changed more than in other ones. So the block direction obtained by this method can correspond to the tendency of the lines in the whole image precisely. In addition, a low-pass filter was used to revise the direction in order to deal with the singularities excursion. With the use of the theory of self-adaptation and fussy, several methods of setting gate of binarization were presented. Although the methods were more and more complex, the local adaptability of gray scale and the matching ability of the direction became stronger and stronger. As a result, the gate obtained by these methods got more and more precise. During the lines healing in order to get rid of the false features, a healing method of broken lines in fingerprint image was introduced. Direction average-divided algorithm, window constraint algorithm and improved window constraint algorithm were designed respectively for the different kind of breaks based on the existing method's merits. These algorithms made the healing process more complete and efficient, were convenient for programming, improved the rid ratios of false ending feature and performed much well on healing the broken lines. A feature extraction method was also proposed, which was based on the coding table of adjacent regions with 8 neighbors. The table about coding and type, ending direction, bifurcation direction, series direction, increment of coordinates in directions were established and made it easy to track the lines'framework to find the features in the fingerprint image. In a word, a lot of fruitful attempts and significant progress on the algorithms of fingerprint identification have been made.