Fingerprint feature reduction by principal Gabor basis function

Fingerprint patterns are full of ridges and valleys and these structures provide essential information for matching, recognition, and classi"cation. Conventionally, most researchers use minutiae, a group of ridge endings and bifurcations, as the features of "ngerprint patterns [1]. Unfortunately, the minutia-based approach contains many time-consumption steps and relies heavily on the quality of input images. In "ngerprint images, however, minutiae are not always clear even though the information of ridge directions and inter-ridge distances is preserved. To avoid above drawbacks, therefore, we proposed a Gabor-based approach to extract features directly from the raw images without involving preprocessing and convolution [2]. This approach applied a bank of Gabor "lters to every region of "ngerprint images and used their responses as the feature vectors for recognition. That is, we did not concern the positions of minutiae, but the responses of a bank of Gabor "lters. To obtain a satisfactory recognition result, nonetheless, this approach needs a lot of time to "nd a suitable bank of Gabor "lters by global consideration. In fact, each local "ngerprint has its particular ridge direction and spatial-frequency. In order to capture these intrinsic characteristics of ridge structures, we propose a local Gabor-based approach to determine the suitable Gabor "lters by using only local information. For further feature reduction, the selected Gabor "lter is mapped to an index of the complete Gabor basis functions (GBFs). At last, we also compare GBFs in the spatial-frequency domain to illustrate the similarities of ridge structures and analyze the feasibility of the proposed method to test a small-scale access control system.