The research of PCA techniques in computer vision dates back to the late 1980s with initial work performed by Sirovich and Kirby at Brown University. Sirovich and Kirby focused on using a computer to recognize cropped versions of human faces and compared their results to that of recognizing full faces. Their system had good results—and more importantly proved the practicality of such a system. Their pioneering success was followed in 1990 when they extended their system of recognizing full faces by doubling their ensemble size through the use of mirror images. This greatly reduced error rate in recognition of test images. Sirovich has continued research and has published as recently as 2000.

While we were unable to locate a copy of the work performed by Turk and Pentland we would be remiss for not referencing it here. Their 1991 variation of Sirovich and Kirby’s PCA system relied on calculating weights associated with eigenpictures (instead of coefficients) and stored these values in a retrievable database. The goal of their system was to successfully determine if a face was present within a given image or not. Interestingly their ensemble not only included “head on” shots but also included different head sizes, orientations, and lighting conditions—making it robust and reliable for several possible types of test images.

Five years later Swets and Weng were able to develop a PCA based system that varied from other contemporary systems (which were dependent upon the differences between images). They instead used discriminating features yielded from discriminant analysis to make classifications. Their research was not only able to recognize the presence of a face, but to also classify the person in the image making it impressive and worth mentioning here.

We finally note and credit the success that Moghaddam and Pentland had by being the first to develop a PCA system that relied upon probabilistic density estimation. This classification method contrasts with the basic Euclidean classification methods used beforehand. Their paper was published in 1997 along with several other publications attempting to create better ways to classify images, each having their own staple of merit and contribution to the realm of PCA research.

The investigation for our inquiry did not require the acquisition of some of these most recent works due to the fact that our system is based extensively on Sirovich and Kirby’s early research which is now considered old. In consideration of our primary investigation we leave further historical research to the interested reader or direct them to the Literature Survey complied by Zhao and others which offers a comprehensive history of PCA studies.