A Finger-Vein Image Quality Assessment Algorithm Combined with Improved SMOTE and Convolutional Neural Network

Biometric identification technology is a technology that uses human biological characteristics for identity identification. Although it is a relatively new technology, it is widely favored for its characteristics that they are difficult to be forged, stable, accurate and non-invasive. For Finger-vein recognition, the distribution and structure of vein branches and curves are random and unique to each individual, so matching algorithms can be used to identify individuals. Finger-vein image quality, as an important guarantee for the accuracy of recognition system, should be paid more attention. The purpose of this project is to evaluate the quality of finger-vein images and improve the accuracy of finger-vein recognition by deciding whether to retain the collected finger-vein images. In this paper, a traditional finger-vein recognition algorithm, called adaptive histogram of competitive Gabor responses, is produced to distinguish the high quality and low quality images. Then, due to the extreme imbalance of the high quality and low quality images, improved SMOTE is used to get the low quality images, and finally, uses the convolutional neural network to differentiate these images.