A KNN Classifier for Face Recognition

Face Recognition has always been a hot topic, especially when the prevalence of Covid-19 calls for ways involving less physical contact in places where personnel identification is critical. However, although there are various algorithms for face recognition, there is limited research in determining the performance of these algorithms using scientific evidence. To address this issue, this paper evaluates the performance of K-Nearest Neighbors (KNN) for face recognition under different situations. To make the result of this study more applicable, this paper aims to train and test the model using photographs taken in profile and partially covered faces to simulate the situation in which the object needs to be identified does not face the camera at a right angle or wears masks. The experimental results demonstrate that K-Nearest Neighbors (KNN) achieved superior performance in recognizing uncovered frontal faces, with a success probability of 95.0%. Nevertheless, the model has a less satisfactory result when classifying profile or masked faces, and the corresponding success probability for the former is 22.2%, the latter 2.22%. It is worth remarking that the accuracy of KNN classifier when used in face recognition is 100% for uncovered frontal faces and 74.7% for covered ones.