Cancelable fusion-based face recognition

Biometric recognition refers to the automated process of recognizing individuals using their biometric patterns. Recent advancements in deep learning and computer vision indicate that generic descriptors which are extracted using convolutional neural networks (CNNs) could represent complex image characteristics. This paper presents a number of cancelable fusion-based face recognition (FR) methods; region-based, multi-biometric and hybrid-features. The former included methods incorporate the use of CNNs to extract deep features (DFs). A fusion network combines the DFs to obtain a discriminative facial descriptor. Cancelabilitiy is provided using bioconvolving as an encryption method. In the region-based method, the DFs are extracted from different face regions. The multi-biometric method uses different biometric traits to train multiple CNNs. The hybrid-features method merges the merits of deep-learned features and hand-crafted features to obtain a more representative output. Also, an efficient CNN model is proposed. Experimental results on various datasets prove that; (a) the proposed CNN model achieves remarkable results compared to other state-of-the-art CNNs, (b) region-based method is superior to multi-biometric and hybrid-features methods and (c) the utilization of bio-convolving method increases the system security with a slight degradation in the recognition accuracy.