Adaptive Frame Selection for Enhanced Face Recognition in Low-Resolution Videos

Adaptive Frame Selection for Enhanced Face Recognition in Low-Resolution Videos by Raghavender Reddy Jillela Master of Science in Electrical Engineering West Virginia University Arun Ross, PhD., Chair Performing face detection and recognition in low-resolution videos (e.g., surveillance videos) is a challenging task. To enhance the biometric content in these videos, imagelevel and score-level fusion techniques can be used to consolidate the information available in successive low-resolution frames. In particular, super-resolution can be used to perform image-level fusion while the simple sum-rule can be used to perform score-level fusion. In this thesis we propose a technique which adaptively selects low-resolution frames for fusion based on optical flow information. The proposed technique automatically disregards frames that may cause severe artifacts in the super-resolved output by examining the optical flow matrices pertaining to successive frames. Experimental results demonstrate an improvement in the identification performance when adaptive frame selection is used to perform super-resolution. In addition, improvements in output image quality and computation time are observed. In score-level fusion, the low-resolution frames are first spatially interpolated and the simple sum rule is used to consolidate the match scores generated using the interpolated frames. On comparing the two fusion methods, it is observed that score-level fusion outperforms image-level fusion. This work highlights the importance of adaptive frame selection in the context of fusion.