Fingerprint identification has long been a fascinating and all-important biometric technique for the detection and prevention of crime, terrorist activities and fraud. Think Crime Scene Investigation: our security systems rely heavily on the usage of fingerprints, alongwith other existent and developing biometrics for identification, crime investigation and access control for authentication and surveillance systems, due to their properties of uniqueness and intransigence. Further, automation of database searching and matching tasks is facilitated through fingerprint classification. Accuracy in identification technologies could eliminate major resource losses through credit card frauds, illegal cellular bandwidth usage, ATM related fraud, etc.
So what are the steps involved in the whole identification process? The architecture mainly consists of a user interface, system database, enrollment and authentication modules. The user interface accepts a fingerprint input. The database consists of a record of authorized entries. The enrollment module enrolls users and their corresponding fingerprints into the database. Authentication is to identify the person accessing the system. There are several distinguishing factors for classifying fingerprints. We consider certain features, such as ridges, and valleys, which are present in the epidermal skin surface. These features form certain geometrical patterns, along with details like ridge divides, endings, crossovers, etc. which are anomalies, and are referred to as minutiae. The directions and locations of the minutiae relative to identifiable ridge or valley structures define the unique print and characterize it. Minutiae form the basis of majority of automated verification systems today.
Classification may alternately be based on singular points, called core point, and delta point. A representative index, called the poincare index is calculated by adding the changes in angle around a curve turning in anticlockwise direction. Depending on a turn of 0o , 180 o or -180 o , the point may be labeled as ordinary, core, or delta. Next, the number of core-delta pairs determines the type of fingerprint class. For example, an arch contains no core-delta pairs. A loop can consist of 1 core-delta pair, a whorl has 2, and so on (Figure 1).
About 6 classes are recognized and each fingerprint may be placed into a class for simplifying the search process in a database consisting of millions of records. Further classifications are possible for each primary class, such as the left loop and right loop categories within loop class. An image matching technique is employed to detect the similarities between a record and an obtained sample. A machine learning algorithm is easily constructed from a graphical classification development. (Fig 3).
Fingerprint based identification algorithms primarily involve feature extraction, enhancement, matching and classification image processing algorithms. The challenges faced in the identification process include transformations in the images with respect to references by way of rotations, translations, elastic distortions and noise factors. There is also a possibility that a highly representative portion of the fingerprint may not have been recorded at all.
However, with ongoing and recent advances in this field of fingerprint sensing technology, as well as matching speed by automatic personal identification systems, more robust and foolproof methods can be devised. Emerging applications, such as fingerprint based smart-cards, can employ fingerprint identification in integration with other biometric/non-biometric techniques to ensure the development of overall, truly mature identification solutions.
