Pose Invariant Face Recognition and Measures of Performance Evaluation

EasyChair Preprint 185

Abstract

Face recognition is one of the important biometrics used now-a-days. This work mainly focuses on recognition of face irrespective of pose and occlusion. Occlusion here means the person has spectacles, different hairstyles etc.. There are various methods to handle above mentioned challenges such as occlusion and various poses. In this paper we will implement four of the methods namely Eigenfaces based on Principal Component Analysis(PCA), Fisherfaces, K-Nearest Neighbors(KNN),Naive Bayes which are mainly used for pose and occlusion invariant face recognition. The results obtained from the above algorithms are compared to assess the good performer in terms of face recognition accuracy. First, Eigenfaces which is based on PCA is implemented. Training set images which include normal front faces of person are provided for learning purpose. Test set images include occluded images such as person wearing spectacles, different hairstyles, slightly tilted in the screen are used to perform testing of face recognition. Next, we implemented Fisherfaces, KNN, Naïve Bayes algorithms for pose invariant face recognition are applied on the same dataset. Now all these algorithms are tested with available test set to evaluate the face recognition accuracy.

Keyphrases: Eigenfaces, Fisherfaces, KNN, Naive Bayes, Principal Component Analysis, face recognition

@booklet{EasyChair:185,
  author    = {S Roja Reddy and N Lalithamani},
  title     = {Pose Invariant Face Recognition and Measures of Performance Evaluation},
  howpublished = {EasyChair Preprint 185},
  year      = {EasyChair, 2018}}