The Face Recognition Performance in Every Lighting Condition of Testing Data

Abstract

Face recognition system is still a topic that is widely studied today, because it has problems that are quite complex. One problem is related to the illumination factor, which significantly influences face recognition performance. One of the methods often used for object recognition problems that is influenced by illumination factors is the obust Regression, which is also used in this research. The problem to be solved is related to the performance of face recognition in various lighting conditions. Using one of the face databases, the Yale Face Database B Cropped, an experiment was performed on each lighting condition of the face data. In the Yale Face Database B Cropped, there are 10 individuals, where each individual has 64 variations of images based on illuminations which are grouped into 5 subsets according to their contrast level. Experiments are generally carried out using all subsets as training data and all other data as testing data, where the results show general face recognition performance (in all lighting conditions). In this research, experiments were carried out for each subset as testing data, where the results showed the performance of face recognition in each lighting condition. The experimental results showed the highest face recognition performance in subset 1 (98.21%), while the lowest performance is in subset 4 (84.46%). The trend of the experimental results shows that the face recognition performance of the Robust Regression method is getting better on face data with a higher level of contrast.