Research On Methods Of Fingerprint Image Segmentation

As an important step of fingerprint image preprocessing, fingerprint segmentation which has a great significance in the system performance can not only improve the accuracy of the feature extraction, but also reduce the time of fingerprint preprocessing. This paper mainly investigates the theory and methods of fingerprint image segmentation, and some of the segmentation methods are classified and discussed. The main works include:(1)A comprehensive review of the fingerprint image segmentation methods is given. Basing on extracting fingerprint features, we first classify the methods of fingerprint image segmentation into two classes, which are the feature fusion segmentation method and multi-level segmentation method. And then, we introduce the two methods in detail.(2)In the aspect of the feature fusion based fingerprint segmentation method, we investigate the Minimum Square Error rule based fingerprint segmentation algorithm and a adaptive texture feature based fingerprint segmentation algorithm. First, we investigate the linear classifier based fingerprint segmentation algorithm, which uses a combination of three variance, mean and ridge orientation features of fingerprint image, and presents the improved algorithm of using the Minimum Square Error rule. The experimental results demonstrate the effectiveness of the improved algorithm, especially in low quality images. Second, we investigate the traditional variance based fingerprint segmentation algorithm, and present a adaptive texture feature based fingerprint segmentation algorithm which combine texture feature and variance. The experimental results demonstrate the effect of the proposed segmentation algorithm is better than the variance based fingerprint segmentation, and it also has stronger resistance to noise. (3)In the aspect of the multi-level based fingerprint segmentation method, we investigate the multi-level segmentation algorithm of fingerprint image. First, to improve the method of calculating the orientation image, the paper uses a better robust method. Second, we present an improved algorithm of setting the thresholds automatically for the disadvantage of setting some thresholds based on experience. The experimental results demonstrate the high accuracy rate of the improved algorithm.