Enhancement Techniques for Improving Facial Recognition Performance in Convolutional Neural Networks

The advent of convolutional neural networks (CNNs) to the development of face recognition system has been a game changer in the field of computer vision and pattern recognition. This research work uses a pre trained MobileNet-V1 model to develop an effective CNN model capable of high performance. We also tackle several common facial recognition challenges which include occlusions, illumination variations, make-ups, pose variation and ageing through the use of several improvement techniques. The techniques include adopting a less computationally costly approach, transfer learning and hyper-parameter finetuning. The Top-1 accuracy 70.6% and Top-5 accuracy 89.5% of the base MobileNet-V1 model has been improved using these techniques to achieve training accuracy of 95% and accuracies of 96.4%, 98.0% and 99.1% on the Pins face recognition dataset, FaceScrub data-set and LFW data-set, respectively. The work done so far illustrates the need for further research into improvement techniques for convolutional neural networks.