Comparing Performance And Computational Efficiency Of Face Recognition Approaches
Eshani Akanksha Bisht, Purnendu Prabhat, Sachin Kumar · 2022
Some of the simplest tasks for a human to accomplish are the most difficult for a machine to solve. Face recognition is an example of this type of problem. Researchers have been attempting to solve face recognition since the 1960s. It's come a long way, and there have been many potential solutions offered. Most of these solutions are provided based on machine learning techniques. The utilisation of edge devices such as smartphones, smartwatches, automotive devices, and other smart home products has skyrocketed the use and development of face recognition in recent years. The main goal of this research paper is to do a thorough investigation and then choose the finest potential solutions for a face recognition application. This study's main objective was to evaluate various solutions theoretically and experimentally. It is not possible to experimentally evaluate every face recognition method available in literature. So, we conducted a literature survey and chose 3 of the most accurate models, namely FaceNet, MobileNet and InceptionResNet. The famous five celebrity faces dataset is used to train and test the models on Google Colab platform. Among all the three models, it was found that FaceNet provided the most reliable results in terms of accuracy while being computationally efficient. In the future, other methods and models for face recognition can be investigated and applied. Finding a model that is even more accurate than FaceNet will be an interesting task.