Facial Landmarks Detection and Expression Recognition in the Dark
Abstract
Facial landmark detection has been widely adopted for body language analysis and facial identification task. A variety of facial landmark detectors have been proposed in different approaches, such as AAM, AdaBoost, LBF and DPM. However, most detectors were trained and tested on high resolution images with controlled environments. Recent study has focused on robust landmark detectors and obtained increasing excellent performance under different poses and light conditions. However, it remains an open question about implementing facial landmark detection in extremely dark images. Our implementation is to build an application for facial expression analysis in extremely dark environments by landmarks. To address this problem, we explored different dark image enhancement methods to facilitate landmark detection. And we designed landmark correct- ness methods to evaluate landmarks’ localization. This step guarantees the accuracy of expression recognition. Then, we analyzed the feature extraction methods, such as HOG, polar coordinate and landmarks’ distance, and normalization methods for facial expression recognition. Compared with the existing facial expression recognition system, our system is more robust in the dark environment, and performs very well in detecting happy and surprising.
Citation
Wang, Qiyu (2020). Facial Landmarks Detection and Expression Recognition in the Dark. Master's thesis, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /192754.