Face recognition is challenging due to the similarity of facial shapes and significant variations in images of the same face under different conditions. This paper investigates how image representations, often considered insensitive to illumination changes, perform in recognizing faces under varying lighting and viewing conditions. A controlled database of faces was used, with each imaging parameter (illumination, viewing position, and expression) controlled separately. The study compared variations between images of the same face with variations between images of different faces. The main finding is that variations in images of the same face due to illumination and viewing directions are typically larger than variations due to face identity. For illumination changes, this is almost complete except for representations emphasizing horizontal features. Even for these, systems based on such representations fail to recognize up to 30% of faces in the database. The study concludes that these representations are insufficient to overcome variations due to illumination, viewing position, and expression. Several image representations were tested, including edge maps, Gabor filters, and derivatives of grey-level images. These were processed with log functions to reduce sensitivity to illumination. However, even with these methods, recognition rates remained low, with miss-percentage (percentage of faces not recognized) ranging from 30% to 100%. For viewing position, miss-percentage was above 84%, indicating poor recognition. For expression, miss-percentage varied depending on the face region considered, with higher rates for partial face regions. The study highlights the difficulty of recognizing faces under varying lighting and viewing conditions. It suggests that existing representations are insufficient and that further research is needed to develop more robust methods. The paper also discusses potential approaches, including multiple image models and model-based methods, but emphasizes the need for representations that are insensitive to illumination and viewing position. The results indicate that while some representations reduce the impact of illumination, they still fail to match human performance in face recognition. The study concludes that overcoming illumination variations is a fundamental challenge in face recognition, requiring further research into domain-specific representations.Face recognition is challenging due to the similarity of facial shapes and significant variations in images of the same face under different conditions. This paper investigates how image representations, often considered insensitive to illumination changes, perform in recognizing faces under varying lighting and viewing conditions. A controlled database of faces was used, with each imaging parameter (illumination, viewing position, and expression) controlled separately. The study compared variations between images of the same face with variations between images of different faces. The main finding is that variations in images of the same face due to illumination and viewing directions are typically larger than variations due to face identity. For illumination changes, this is almost complete except for representations emphasizing horizontal features. Even for these, systems based on such representations fail to recognize up to 30% of faces in the database. The study concludes that these representations are insufficient to overcome variations due to illumination, viewing position, and expression. Several image representations were tested, including edge maps, Gabor filters, and derivatives of grey-level images. These were processed with log functions to reduce sensitivity to illumination. However, even with these methods, recognition rates remained low, with miss-percentage (percentage of faces not recognized) ranging from 30% to 100%. For viewing position, miss-percentage was above 84%, indicating poor recognition. For expression, miss-percentage varied depending on the face region considered, with higher rates for partial face regions. The study highlights the difficulty of recognizing faces under varying lighting and viewing conditions. It suggests that existing representations are insufficient and that further research is needed to develop more robust methods. The paper also discusses potential approaches, including multiple image models and model-based methods, but emphasizes the need for representations that are insensitive to illumination and viewing position. The results indicate that while some representations reduce the impact of illumination, they still fail to match human performance in face recognition. The study concludes that overcoming illumination variations is a fundamental challenge in face recognition, requiring further research into domain-specific representations.