Unsupervised learning dimensionality reduction algorithm PCA for face recognition

Some challenges of developing face recognition system is to train examples of different poses, illuminations, contrast and background conditions. Even though if the system works okay with the tested data set; it fails big time to perform accurately with new data set with different attributes. This manuscript is focusing on the mentioned problem statement by deploying PCA and implementing the concept of splitting the dataset in training, cross-validation, and test set to compute misclassification error. It enables the algorithm to be a better fit to be used for the new training set. Principal component analysis (PCA) popularly known as dimensionality reduction algorithm often referred as black box, which is used widely but poorly understood. The hope with this manuscript is to provide roadmap for a deep understanding the PCA and mathematics of its step by step process to develop the desired recognition system. The interesting part of this manuscript is the selection of database. The source of images were databases of AT&T labs, CVL Face Database, Face 94, Essex database and real times images taken by us with a standard professional camera. The images included in the database are of 10,000 people from various racial origins, it spans evenly over the age group that is from 18 to 60 years old. All the experiments were executed on MATLAB and the results were obtained with fair accuracies which are discussed at the end.