Quality assessment of facial images for face recognition

• Main Author: Li, Hui.
• Other Authors: Cho Siu-Yeung David
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/36265
• Institution: Nanyang Technological University

In recent years, face recognition has received substantial attentions from both research communities and the commercial, but still remains challenging in real applications. One of the problems faced by facial recognition system in real-time applications is the facial image with poor quality conditions affecting its performance, especially in the face detection and face identification phases. Therefore, the FIQAS (Facial Image Quality Assessment System) is introduced as a pre-processing module, to measure and even improve the quality of the facial images to support and enhance the facial recognition system. In this project, I propose an approach for developing a no-reference FIQAS, which does not require any reference during the assessment, to evaluate the quality of grayscale facial images based on three main quality metrics—the image sharpness, brightness and also the head pose. An image improvement module is also included. The algorithms for assessing each quality metric are implemented with Matlab. A GUI (Graphic User Interface) is also designed for the user to access the system. In the algorithm design, the sharpness of the image is measured upon the spatial variance between the input image and the degraded image generated from it. The brightness evaluation includes average brightness calculation and also the analysis of different lighting conditions by separating the image into different regions. A novel algorithm is proposed for the head pose estimation, which is also based on the symmetric property of facial image. It compares the pattern of the left side with the pattern of the mirror-reversed right side to estimate the head rotation angle. The experimental results are provided to illustrate the performances of these assessment algorithms and also prove the effectiveness of the improvement module.