The paper by Yael Moses, Yael Adini, and Shimon Ullman explores the challenge of face recognition, particularly in compensating for changes in illumination direction. The authors study several image representations, including edge maps, derivatives of the grey-level image, and images convolved with Gabor filters, to assess their sensitivity to illumination changes. Using a controlled database of faces, they compare the differences between images of the same face under different illumination conditions with those between images of distinct faces. The results show that variations due to illumination and viewing directions are often larger than those due to changes in face identity. Despite some representations emphasizing horizontal features, systems based on these representations still fail to recognize up to 30% of the faces in the database. The authors conclude that these representations are insufficient for overcoming variations due to illumination direction alone, and suggest that further research is needed to develop more effective representations. They also discuss potential approaches to address this issue, including multiple image approaches, model-based methods, and class-based methods, and compare the performance of these methods to that of the human visual system.The paper by Yael Moses, Yael Adini, and Shimon Ullman explores the challenge of face recognition, particularly in compensating for changes in illumination direction. The authors study several image representations, including edge maps, derivatives of the grey-level image, and images convolved with Gabor filters, to assess their sensitivity to illumination changes. Using a controlled database of faces, they compare the differences between images of the same face under different illumination conditions with those between images of distinct faces. The results show that variations due to illumination and viewing directions are often larger than those due to changes in face identity. Despite some representations emphasizing horizontal features, systems based on these representations still fail to recognize up to 30% of the faces in the database. The authors conclude that these representations are insufficient for overcoming variations due to illumination direction alone, and suggest that further research is needed to develop more effective representations. They also discuss potential approaches to address this issue, including multiple image approaches, model-based methods, and class-based methods, and compare the performance of these methods to that of the human visual system.