Face classification using PCA and K-nearest neighbor method
Liang, Wen-Wei (2014) Face classification using PCA and K-nearest neighbor method. [Dissertation (University of Nottingham only)]
Principle Component Analysis is one of the most useful method using for human face recognition, our aim in this research is to implement an understandable recognition program with language R; furthermore, comparing K-Nearest Neighbor classifier with two different distance measurement and Support Vector Machines on their efficiency. PCA indeed reduce the run time of processing massive face images data and reach an ideal accuracy in controlled circumstances. We found that Euclidean K-NN classifier generally has better accuracy than Manhattan K-NN classifier; however, it suggested that they are suitable to use for smaller k value, especially one. The eigenfaces applied to construct the face space is essential as well, we reached best accuracies when selecting around 10% to 20% of eigens. However, the reconstruction suggests in an opposite way, more eigenfaces help the rebuilt much ideally; more than half of eigenvectors selection can reconstruct the face to be easily recognised as the original people.
Actions (Archive Staff Only)