Research Of Quality Evaluation Methods On Fingerprint Images

Automatic fingerprint identification system (AFIS), which consists of fingerprint image enhancement, fingerprint classification, feature extraction and fingerprint matching, has drawn a substantial attention in recent years. Although many researchers have done lots of studies on AFIS, it still not satisfying for many reasons. One of the most difficult problems is the identification of low quality fingerprint images. If the quality of fingerprint images is too low, such as dry and wet, will has large affect on the performance of AFIS.In this dissertation, we deeply analyze and investigate the above problems of automatic fingerprint identification system. According to the inadequacies of recent methods in fingerprint quality classification, we proposed an idea which use support vector machine into fingerprint quality classification. We implemented a classification algorithm based on support vector machine. First, we use exit quality evaluation indexes to evaluate fingerprint images. Second, a support vector machine was trained based on the values of these quality indexes. Finally, classify fingerprint images with the trained support vector machine. The result on the fingerprint database proves the effectiveness of the classification algorithm.In the procedure of classification, we found there is a class imbalance problem in the fingerprint database. In order to solve the affect of this problem, we proposed an improved quality classification method based on support vector machine. Before we train the SVM, we use SMOTE algorithm to adjust the proportion of the sample fingerprint images. Then we use the adjusted training set to train the SVM and classify the fingerprint images. The result on the fingerprint database also proves the effectiveness of the improved classification algorithm.It is true that the quality of a fingerprint image has great affect on the extraction of direction information. The lower the quality of a fingerprint image is, the more error of direction information we will get. We know that the real direction field is continuous. However, the wrong direction we get due to the low quality is clutter and discontinuous. Based on this fact, we proposed a new quality evaluation index based on direction field information. The effectiveness of this new index was proved on the test of the fingerprint database.The thesis is organized as follows: Chapter 1 is introduction, which introduces the general institutions and difficulties of fingerprint identification. Chapter 2 presents a new quality classification method based on support vector machine. Chapter 3 introduces an improved method of the algorithm we proposed in chapter 2. A new method of quality classification based on direction field is reported in Chapter 4. Chapter 5 is conclusion and discussion.