Research On Feature Extraction Approaches For Fingerprint Classification

The simple approach is extracting the ridge flows and the number/location information of singular points of fingerprints in fingerprint classification works. Due to the affection of the image quality of fingerprints, these information can’t be achieved accurately. For the reasons above, extracting the orientation and singular points accurately, or extracting other information which can describe the type of fingerprints accurately and robustly is becoming the key step of fingerprint classification works.This paper focuses on the Feature Extraction process of fingerprint classification works, and aims to obtain features which are able to describe the classes of fingerprints, in order to achieve more accurate classification results. The main contents of this article include:(1) Inspired by existing literature a new algorithm of fingerprint classification named as Large-scale Orientation Field Descriptor is proposed in this paper. The algorithm describes approximate orientation pattern near the core point by extracting the direction of the nodes belong to a large-scale annular mesh structure surrounding the core point as feature vector. Because of the simple and efficient feature extraction,experiments show that comparing FingerCode proposed method achieves similar classification accuracy with 20 times computation speed.Adaptive Orientation Field Descriptor, which is the improvment during experiments, adjust the radius of the structure through the effectiveness of the nodes,in order to adapted different scale of fingerprint images. Experiments show that through this feature the robustness to different fingerprint databases is improved with a stable classification accurate.(2) Anew fingerprint classification approach based on DCT Orientation Field is proposed in this paper for reducing the effect of the preprocessing steps. The proposed approach segments fingerprint images dynamically and extracts the Low-frequency parts of the matrix which is the discrete cosine transformation of the orientation field as feature factors to describing approximate orientation pattern of whole fingerprint and describe the class of the fingerprint. Experiments show that the proposed method achieves satisfactory classification accuracy. A multistage classification algorithm is proposed to solve the mis-classification cases in experiments,which improve the accurate of classification through higher ratio of penetration in the database.