Face Recognition Using an Innovative Bag of Features and Support Vector Machines
Discipline: Computer Sciences and Information Management
Subcategory: Computer Science & Information Systems
Session: 2
Room: Park Tower 8219
Jordan T. Williams - Fort Valley State University
Co-Author(s): Dequan E. Medina, Dytalyan Z. Holmes, and Chunhua Dong , Fort Valley State University, GA
Face recognition is a popular area in the research and applications of artificial intelligence. The purpose of face recognition is to find the feature representation of a facial image and recognize the identity of the person present in the image. In this study, a face recognition algorithm is proposed based on the appearance and shape feature representation via Bag of Features (BoF) and classification via support vector machine (SVM). We first extract the appearance and shape features of the images. For appearance information, we use the Gabor filter to extract the facial appearance changes as a set of coefficients on the whole face image. For shape and texture extraction, Histogram of Orientated Gradient (HOG) and Local Binary Pattern (LBP) descriptors are used to obtain the spatial distribution of edges. Next, we build a BoF representation for each training face image using the extracted three features which have proven to yield decent classification accuracy. Finally, the SVM with RBF kernels is trained using a set of training face data and a test face can be classified by the trained SVM. Compared to using the original HOG, LBP and Gabor features for SVM classification, the BoF features achieve a higher accuracy of face recognition.
Funder Acknowledgement(s): Research was sponsored by the Army Research Office and was accomplished under Grant Number W911NF-18-1-0457
Faculty Advisor: Xiangyan Zeng, zengx@fvsu.edu
Role: I conducted the research on feature extraction, including LBP, HOG, and Gabor filters.