Fingerprint Pore Descriptor

Fingerprint matching is of key importance in the automatic recognition system. In order to match two images, a set of matching points between the two images are obtained. Also the points are used to align the matched pattern of the two images which is done mostly using minutiae points.  To overcome the limitation of the traditional fingerprint matching system which uses minutiae location for pore matching, we have used a direct pore matching method which matches pore using pore points (location). The overall pore matching process is divided into two parts, feature vector estimation (fingerprint pore descriptor) and Direct pore matching using two metrics SAD (Sum of Absolute Differences) and SSD (Sum of Squared Differences). 

Pore Descriptor

The extracted pores obtained from the proposed two fingerprint pore model are recorded by their locations. Thus for matching the two input images we have the two set of location of pores (coordinates). Each pore is associated with the descriptor often called as feature vector. In the literature of computer vision a lots of pore descriptor have been proposed to describe point feature [3].Most of them are based on gradients in the local neighborhood to the points. But the gradients in the local neighborhood are not very distinctive.  

Q.Zhao et.al [1], [2] proposed a method for directly building the descriptor from each pores without involving the gradients. The descriptor is built from the pixel value in the local neighborhood to the pore. First of all the fingerprint image is smoothen using the Gaussian filter so as to remove the noise in the fingerprint. Then a circular neighborhood is set up to each pore. Nitin et al. [4] proposed the different shaped based cropping such as circular cropping, triangular cropping. The circular neighborhood is rotated such that the ridge orientation at the locations of the pore becomes horizontal. By using this rotated circular neighborhood, the descriptor is rotation invariant. A feature vector is the obtained by flattening the neighborhood and it is normalized to have zero mean and unit length. 

The normalization makes the feature vector invariant to monotone contrast changes and simplifies the substituent computation .The flattening is nothing but converting the two dimensional block to one dimensional block. This feature vector is defined as the descriptor for the pore. Fig.1 shows the basic steps required for building a descriptor across each pore obtained from pore extraction. This feature vector obtaining process is performed for both the two input image so as to obtain the two pore descriptor which are to be matched. Fig.2 shows the two pore descriptor obtained from two different images which are to be matched. This descriptor obtained around the each pore neighborhood will be compared to find the most similar match.

Fig.1: Flowchart of constructing the descriptor for a pore marked by circle

Fig.2: Two Fingerprint Pore Descriptor

 Thus the above two fingerprint pore descriptors are used for one to one fingerprint matching.In the next post i will discuss the two metrics SAD and SSD used to match this feature points or pore descriptor.

References:

[1] Q. Zhao, L. Zhang, D. Zhang, N. Luo, “Direct Pore Matching for Fingerprint Recognition”, IAPR/IEEE 3^rd International Conference on Biometrics (ICB2009), Italy, pp. 1050-1061, 2010.

[2]   Q. Zhao, L. Zhang, D. Zhang, N. Luo, “Fingerprint Pore Matching based on Sparse Representation” International Conference on Pattern Recognition, 1051-4651,2010.

[3]   K. Mikolajczyk, C. Schmid: A performance Evaluation of Local Descriptors, IEEE Trans. Pattern Analysis and machine Intelligence 27(10), 1615-1630 (2005).

[4]   N. Saluja, A. Kumar, Amisha and R. Khanna. “Cropping image in rectangular, circular, square and triangular form using matlab”, National Conference on Computational Instrumentation, Chandigarh, India, March 2010.