Identification with CNNs by Fusing Face, Fingerprint, and Voice Recognition
Samuel B. Ehgartner, Ismail I. Jouny · 2023
In this paper, classifications using three convolutional neural networks (CNN) are fused together to explore the performance of an identification system. In this system facial, voice, and fingerprint recognition are used to demonstrate fusion. These three sets of data vary in complexity and represent a range of image quality and perspectives. A convolutional neural network’s success is entirely dependent upon its ability to accurately classify images. By fusing CNNs, the algorithm has more data to add in the decision-making process, thereby increasing classification accuracy. In order to demonstrate fusion’s potential advantages, a mock security system is designed to evaluate reliability. Three methods of fusion are compared to determine the most effective way to boost performance of the designed fusion architecture. Each method is evaluated on 50 unique profiles containing face, fingerprint and voice data. To further determine fusion’s effectiveness, noise was added to an identical dataset and evaluated using the same methods.