Finger vein and hand-dorsal vein multi-modal biometric system based on convolution neural network

Abstract

Recently, the increase in the criminal in cyberspace due to unauthorized actions by cyber-attacks on identification systems. People are searching for more accurate methods of personal identification with growing numbers. However, Multimodal biometric systems are currently being studied extensively for personal identification and verification. When the diversity of features is increasing merging with deep convolutional neural networks, then the security of biometrics systems is improved. Beside that CNN is heavily reliant on a huge data set to avoid overfitting problem. Unfortunately, many application domains do not have access to big data. This paper used two combination technique methods of the fusion of finger veins and hand-dorsal vein biometric traits and was implemented. These methods were the inspirations for the expansion technique. The expansion process used to increase the number of training images and thus trains the network of the largest number of features to overcome overfitting. Thus, the CNN model was proposed for training and classify the resulted fusion images. The proposed system tested on two public data sets (Yilong Yin and Badawi) and the experimental results proved that the system was reliable .The first fusion technique which combines the two images separately (without merging the pixels) gave an accuracy 96%. The second method fused all two opposite pixels into one pixel and gave accuracy 97% .Finally, when the second method with expansion gave an accuracy 100%.