Discipline: Computer Sciences and Information Management
Subcategory: Computer Science & Information Systems
Session: 3
Room: Hoover
Dytalyan Holmes - Fort Valley State University
Co-Author(s): Dr. Chunhua Dong, Fort Valley State University, Fort Valley, GA
Face recognition is a popular area of research in the applications of artificial intelligence. Accurate detection of region of interest (ROI) is a key step in a face recognition system. Viola-Jones is an object detection framework primarily for face detection. However, the performance of the Viola-Jones algorithm may suffer from missed faces and wrongly detected non-face objects. To eliminate the non-face objects and improve the face detection performance, we propose to incorporate a cross-object comparison technique into the Viola-Jones framework. Firstly, the face objects detected by the Viola-Jones framework are cropped and resized into a common dimension. Next, the feature descriptor of histogram of oriented gradients (HOG) is extracted for each object. The HOG features provide unique texture information about the image, which helps achieve more accurate face detection. Finally, a discriminative weight is calculated for each object using a validation comparison process. Non-face objects are then removed based on these discriminative weights. Experimental results show that the proposed scheme can effectively eliminate the non-face objects and thus achieves a higher accuracy of face detection than the classical Viola-Jones method. Our future work is to reduce the number of missed face objects in the Viola-Jones algorithm.
Funder Acknowledgement(s): Army Research Office Grant Number W911NF-18-1-0457
Faculty Advisor: Dr. Xiangyan Zeng, zengx@fvsu.edu
Role: Wrote code, researched different algorithms, and did trial and error with different algorithms.