Exploring the Trade-Off Between Performance and Cost in Facial Recognition: Deep Learning Versus Traditional Computer Vision

During the late 20th century, the rise of computers and technology led to collaborations between mathematicians and computer scientists to develop systems that could solve various problems using traditional approaches. One such problem was facial recognition, which enables the identification of human faces without interference. Traditional machine learning methods were used to build these systems until recently, when deep neural networks trained with massive datasets replaced them. Deep learning has pushed the boundaries of what is achievable in facial recognition. However, traditional computer vision methods, which had been improving for years prior to the emergence of deep learning, still have room for improvement. This study aims to compare the accuracy of facial recognition algorithms, specifically FaceNet and Eigenfaces, from deep learning and traditional computer vision domains. The availability of physical machine power and dataset scale creates a tradeoff between performance and cost. Traditional computer vision techniques may be suitable for low-power applications, while deep neural networks provide better accuracy results when high-powered machines are available. The performance benefits of deep learning outweigh the data scale and training execution time required.