Optimized Deep Learning Model for Pose and Expression Invariant Face Recognition in an IoT-Cloud Environment

Abstract

Face recognition from massive data in Internet-of-Things (IoT)-cloud environments is challenging with limitations in learning the pose and facial expression variations. An intelligent Pose and Expression Invariant Face Recognition Model (PEIFRM) is developed in this paper by extracting the local features of face pose and expression variations using Temporal Stacked Convolutional Denoising Autoencoder (TSCDAE) and Optimized Siamese Convolutional Ladder Networks (OSCLN) for recognition. TSCDAE acquires the local informative features of persons' facial pose and expression variations through different color components. OSCLN is an integration of Siamese neural networks (SNN), semi-supervised ladder form of convolutional neural networks (CNN) and Artificial Lizard Search Optimization (ALSO) to improve the training speed and reduce the error rate with local and global feature fusion to improve the recognition accuracy. Experimental results comparisons showed that the proposed PEIFRM model achieved 98.95% and 99.75% accuracies for LFW and ORL datasets, respectively.