International Journal of Science, Engineering and TechnologyAuthors: Ritik Singh, Sri Saumya, Sameer Khan
Abstract: – Currently, one of the newest areas of study in biometric recognition is finger vein recognition. Although the Gabor filter's settings are hard to modify, it has been widely employed for vein and finger recognition. Here, an adaptive-learning Gabor filter is proposed to address this issue. Based on the goal function, the gradient of the Gabor-filter parameters is calculated, we merge convolutional neural networks with a Gabor filter. Then, we optimize its parameters by back-propagation. The Gabor filter's θ parameter can be learned at same angle as the vein texture in an image of a finger vein. There is a relationship between the Gabor filter's σ and λ parameters, and the latter can converge to ideal value. With this method, we not only select appropriate and effective Gabor filter parameters for filter bank construction, but we also consider the interrelationships between those parameters. Lastly, we conduct tests on four publicly available finger vein datasets. According to experimental results, our approach performs better in finger vein classification than the most advanced techniques.
DOI: http://doi.org/
