An unsupervised face recognition model by self-enhanced side information

In this thesis, a new unsupervised face recognition model has been proposed in which the face images are transformed through the identity filter banks and the new subimages derived from original image are used as positive side-information in Relevant Component Analysis (RCA). This model has ran over ORL, and AR face databases by using Cosinus, Correlation based, Squared Euclid, Manhattan and Mahalonobis distance metrics and a significant classification improvement is observed.By subsampling with identity filter banks four new subimages of a quarter size of the original image are obtained and the number of useful eigenvectors are increased which will help to improve the accurate classification rates. While in classical techniques the best accurate recognition rate and incorrect recognition rate are standing in a small eigenvalue space; with the help of big amount of useful eigenvectors, this space is extended to a large scale of surface where the data can be properly spreaded.In this new model RCA, followed after PCA, has leaded us to reduce the weight of irrelevant components in projection space and improve the accurate classification performance up to 12.69 %.