Feature Extraction and Feature Selection Techniques for Face Recognition

Abstract

Face recognition is one amongst the most significant applications of image processing. It’s a challenge to make an automated system which will be comparable to human ability for recognizing faces. Although other biometric techniques are reliable but problem with these techniques is that the individual has to interact with the system, while face recognition is a non-intrusive technique which can be performed without interaction with the system.

Feature extraction is most crucial part of any pattern recognition problem. In our work we have applied different techniques for facial feature extraction like DWT, LWT, and Fast Discrete Curvelet Transform. For dimensionality reduction PCA was applied and Linear-SVM based classifier is used for classification. It was found in the study that feature extracted based on Curvelet transform were giving better results. Extracted feature were also tested for different feature selection algorithms. It was found that Conditional Redundancy based selected feature in face recognition outperforms other feature selection algorithms.