A neural-based minutiae pair identification method for touch-less fingerprint images

Contact-based sensors are the traditional devices used to capture fingerprint images in commercial and homeland security applications. Contact-less systems achieve the fingerprint capture by vision systems avoiding that users touch any parts of the biometric device. Typically, the finger is placed in the working area of an optics system coupled with a CCD module. The captured light pattern on the finger is related to the real ridges and valleys of the user fingertip, but the obtained images present important differences from the traditional fingerprint images. These differences are related to multiple factors such as light, focus, blur, and the color of the skin. Unfortunately, the identity comparison methods designed for fingerprint images captured with touch-based sensors do not obtain sufficient accuracy when are directly applied to touch-less images. Recent works show that multiple views analysis and 3D reconstruction can enhance the final biometric accuracy of such systems. In this paper we propose a new method for the identification of the minutiae pairs between two views of the same finger, an important step in the 3D reconstruction of the fingerprint template. The method is divisible in the sequent tasks: first, an image preprocessing step is performed; second, a set of candidate minutiae pairs is selected in the two images, then a list of candidate pairs is created; last, a set of local features centered around the two minutiae is produced and processed by a classifier based on a trained neural network. The output of the system is the list of the minutiae pairs present in the input images. Experiments show that the method is feasible and accurate in different light conditions and setup configurations.