Researches On The Key Technologies In Incomplete Fingerprint Recognition

With human society stepping into the information era, the requirement of authentication is becoming increasingly urgent. Therefore, the research over biometric recognition is in full swing and has broad prospect. Because of its uniqueness and immutability, fingerprint recognition has become the most effective way applied in identification recognition and authentication. With high practicality and feasibility, fingerprint recognition had been developing rapidly especially since19century, and it has already became far more common and has force of law.In recent years, many academies and industries have been making an in-depth research on the key technologies of incomplete fingerprint recognition, and the proposed algorithms perform well both in accuracy and efficiency on high quality fingerprint recognition. But it is still challenges for the low quality incomplete fingerprint recognition because of feature loss and non-linear deformation caused by stain, scar and broken on fingerprint images. Meanwhile, high quality fingerprints only occupy a small portion in fingerprint files of our country, and incomplete fingerprints occupy a certain proportion. Therefore, this problem needs to be solved urgently by pay more attentions to recognize incomplete fingerprint accurately and effectively.This thesis has done many researches on the key technologies of incomplete fingerprint recognition, including image enhancement, feature extraction, fingerprint matching and fingerprint indexing, and makes the following contributions:1. The algorithm of incomplete fingerprint regions reconstruction and reparation is proposed based on information entropy. Since fingerprint image is the original input data of fingerprint recognition, the quality of image directly affects the precision of extracted feature, and then affects the recognition accuracy. Due to low quality circumstance such as mess, broken and blur of ridges in incomplete fingerprint, it is difficult to extract reliable minutiae. Therefore, there is first of all to do fingerprint enhancement for incomplete fingerprint, especially do incomplete regions reconstruction and reparation. Our research combines minutiae and orientation field to thoroughly estimate the unknown orientation field of incomplete regions. Then, previous outcome is used to analyze distribution of ridges and minutiae, and get several reconstruction schemes. Furthermore, schemes are measured by using entropy and the best is chosen. The proposed algorithm is proved to be effective in reconstructing and repairing incomplete fingerprint, and hence improves the performance of matching and indexing.2. The algorithm of fingerprint matching is proposed based on fusion feature and pattern entropy.Generally, fingerprint matching consists of two steps:feature extraction and similarity measurement. Firstly, feature extraction directly affects the result of recognition. Traditional methods of feature extraction depend on the reliability of information in fingerprint image to a great extent, but false features usually exist due to low quality of incomplete fingerprint image. Secondly, after feature extraction, similarity measurement is to measure the similarity of two feature sets and then to decide whether two fingerprints are similar or not. Most of the traditional methods of similarity measurement are sensitive to errors of location and orientation of feature sets, and also cannot eliminate false matches. Our research proposes a fusion feature with more comprehensive discrimination based on both the minutiae and orientation field feature. Furthermore, we use pattern entropy for similarity measure which could measure the consistency of feature sets and eliminate most of the false matches. Finally, we got a satisfactory accuracy of incomplete fingerprint matching in the experiment. 3. The algorithm of fingerprint matching based on improved GA-PSO (Generitic Algorithm-Particle Swarm Optimization).Minutiae-based matching algorithm is most commonly applied in fingerprint matching. It has the advantages of low storage cost, simple operation and great results for high quality fingerprint. However, traditional minutiae-based matching algorithm has its deficiencies in incomplete fingerprint matching. For instance, it requires more in location of local feature sets and reliability of pre-alignment. Therefore, it is hard to ensure the matching result. Our research improves minutiae-based matching based on optimization algorithms. Firstly, population initialization is improved by using optimal solution based on pre-alignment as parts of original individuals. This could ensure both the revolution direction and the random search. Secondly, fitness function is improved by using fusion feature based on minutiae and orientation field. This could remedy the defects of few and low reliability minutiae extracted in incomplete fingerprint. Furthermore, improved GA-PSO is proposed which is a complementary scheme for solely GA or PSO. It could overcome the redundancy iteration in GA and promote the diversity of the solutions.4. The algorithm of fingerprint indexing based on BMHash (b-bit Minwise Hashing).Identifying an unknown fingerprint over a large-scale database still faces challenging problems in terms of both efficiency and usability. To reduce the search space and increase the recognition efficiency, fingerprint indexing is often adopted as a pre-filtering technique to select several most similar fingerprints with query fingerprint. Ideally, fingerprint indexing algorithm could do indexing and matching rapidly and steadily. But it is difficult for incomplete fingerprint indexing because of differences and noise of images. Our research constructs the indexing feature vectors from the minutiae triplets feature by clustering. It could reduce computational burden greatly. And then, BMHash is used to generate index value, which proved to be very effective in mapping mass feature vectors into a small hash table. In this way, we could make a tradeoff between space and time, speed and accuracy.