Thermal and Visual Face Recognition using Eigenfaces and Transfer Learning

Images with dark backgrounds captured using thermal camera sensors are used for thermal face recognition, whereas a visual camera captures images in RGB format and can be used with good lighting conditions. The light conditions do not affect the thermal images as it entirely depends on the heat patterns which are emitted by the body. Human faces and bodies have nearly the same temperature, which varies from 35.5 °C to 37.5°C. The study on face recognition started in 1960. Individual faces are used for authentication into various systems throughout the world. This paper combines the traditional approach of face recognition using Eigenfaces and the transfer learning approach. In the process, first, the face is cropped out using a deep learning model. Then the features are extracted by using a pre-trained deep learning model, which essentially converts an image to a low dimensional array. Eigenfaces are formed using the array for each person and when new face images are given for recognition, the system calculates the Euclidean distance between the new image eigenface to all previously stored eigenfaces. The solution is tested on the IRIS Thermal and Visual face dataset and accuracy is found to be 62.9% and 91.93% respectively.