Deep feature extraction and classification for finger vein images

Abstract

Cross-sectoral insecurity increased crime, and piracy is all major topics these days. Fur- thermore, people’s movement, financial transactions, and access to services necessitate an urgent need to guarantee their identity. Traditional security solutions rely on pre- viously learned information (PIN codes, passwords) or access tokens (keys, identifiers, badges). However, in many situations, these technologies are less reliable because they are unable to discern between legitimate persons and scammers. In this master’s thesis, we chose a deep learning finger vein recognition system. This system is difficult to Fal- sification. There are numerous benefits, including ease of use and inexpensive cost. Our work can be divided into two stages. To start, data augmentation utilizing various geo- metrical techniques is used to compensate for the paucity of training samples required for the deep learning model’s training. Second, the four CNN algorithms are used to exe- cute feature extraction and classification tasks in order to validate the person’s identity. The suggested model’s performance is tested and evaluated using the SDUMLA dataset. With Vgg16 and 97.22 percent with Vgg19, 90 percent with the inception model, and 95 percent with MobilenetV2, our suggested technique for the SDUMLA database achieved an accuracy of 93 percent with Vgg16 and 97.22 percent with Vgg19, 90 percent with the inception model, and 95 percent with MobilenetV2. The proposed work achieves good performance when compared to existing methods, according to the findings of the experiments.