On the Extension of the Haar Cascade Algorithm for Face Recognition: Case Study and Results

In this paper, we build upon existing work and suggest an extension of the Haar Cascade face recognition algorithm in order to improve its accuracy and efficiency under varying lighting conditions. For doing so, we add a phase-preserving denoising method, which leads to a faster convergence of the classifier and improved face recognition efficiency. We evaluate the suggested approach by comparing the native Haar Cascade algorithm with our suggestion in terms of efficiency and accuracy within six different options for video filters, depending on varying lighting conditions i.e. broad daylight vs. dark environments. Analysis of results revealed that the suggested approach outperformed the native implementation in most evaluation scenarios and that the best filter options in broad daylight settings would be the No and the Gray filters, whereas in dark environments, the Gray and the Redish filters would be the best options. The Gray filter achieved top ratings in both lighting condition settings yet requires a higher computational cost for facial recognition.