A Comparative Study of Thermal Face Recognition Based on Haar Wavelet Transform (HWT) and Histogram of Gradient (HoG)

A comprehensive evaluation of the visual features applied to the proposed face recognition method on thermal images is presented. Comparative performance analysis is carried out on two different features which includes Histogram of Gradients (HoG), and Haar Wavelet Transform (HWT) plus Principle Component Analysis (PCA), resulting two different classifiers which include Support Vector Machine (SVM) and Multi-Layer Perception (MLP) is conducted. The comparisons are made for the following cases. Case-1 feature extraction (HoG) and classify (SVM), Case-2 feature extraction (HoG) and classify (MLP), Case-3 feature extraction (HWT & PCA) and classify (SVM), and Case-4 feature extraction (HWT & PCA) and classify (MLP). The face recognition experiments were carried out and compared by using python programming. The image dataset has 4400 images obtained from 15 subjects of various ages and genders recorded by the IR-BF503 thermal camera with several head poses, occlusions and illuminations. The performance of the face recognition method is measured based on the MSE, PNSR, bit rate, ratio, and time compression, accuracy, precision, recall, and F1 score. The purpose of this method comparison is to find a suitable method for facial recognition which is used to measure body temperature and heart rate. Finally, the algorithm implemented to case-4 showed a good performance due to its highest accuracy, F1 score, and recall.