Performance Comparison of Various Features for Human Face Recognition using Machine Learning

Facial features are one of the most vital biometrics and are used to identify an individual. Facial recognition is a technology having the capacity to distinguish a specific individual. This technology mainly concentrates on machine learning techniques to learn, acquire, store, and examine facial features to fit them with a database. In this project, the features are extracted in transform domain. Discrete Wavelet Transform (DWT) is applied on images. In transform domain, features like mean, energy, Histogram of Oriented Gradients (HOG), Local Binary Pattern (LBP), Gabor Filters are explored. Appropriate features are extracted from LL, LH, HL, and HH bands to build a machine learning model. Database is divided into training and testing. Model is built based on the 80% of images from the database. Model is tested with 20% images of test data. Three important machine learning algorithms are popular. These are Decision Tree (DT), Support Vector Machines and Naive Bayes (NB). NB is used for probability-based inferences. DT is simpler than SVM. Hence, Decision Tree-based Machine Learning is employed to recognize the faces. Accuracy is used to test the performance of various features.