Due to its mobility and reliability, the outer finger-knuckle-print (FKP) possesses several advantages over other biometric traits of the hand. However, most existing state-of-the-art methods utilize either local features alone or together with global features for FKP verification. These methods often demand high computational cost despite their high verification accuracy. In this paper, we propose a novel and fast matrix projection method for extracting line features from the finger-knuckle-print for person verification. Essentially, both the horizontal and the vertical knuckle lines are extracted by projecting the knuckle print image onto a shift-and-difference matrix. Such a matrix enables directional image shifting and subtraction within a single matrix multiplication. The resultant difference image then goes through a sigmoidal activation for contrast enhancement. Subsequently, the Fourier spectrum of the contrast enhanced image is adopted as the holistic features of the given finger-knuckle-print image. The entire process of extracting the proposed features is expressed in an analytic form to facilitate a fast vectorized implementation. For cognition performance enhancement, the two directional line features are subsequently fused at the score level by minimizing the error counts of the extreme learning machine kernel. Extensive experiments are performed to compare the proposed method with competing methods using three public finger-knuckle-print databases. Our experimental results show encouraging performance in terms of verification accuracy and computational efficiency.
All Science Journal Classification (ASJC) codes
- Computer Vision and Pattern Recognition
- Computer Science Applications
- Cognitive Neuroscience