Much research has been done in the field of automated facial expression recognition because of the importance of facial expressions to understanding human interactions and emotions. While several systems have achieved positive results using either facial model based classification or feature based classification, most of these systems have been tested on subjects in constant lighting conditions. These systems may thus be susceptible to lighting changes since illumination contribute much more to image variation than facial features.
In this report, we augment the BU-4DFE dataset by adding different lighting conditions to 3D images of subjects performing different facial expressions. Then we develop an image processing pipeline to rectify the effects of illumination on the images, hoping to preserve high classification rate even in harsh lighting conditions. Then we test our pipeline on two measurement: classification accuracy based on a LDA model and SIFT keypoint repeatability.
For our results, we found that our image processing pipeline helped improve classification accuracy when performing LDA to identify images in dark lighting conditions. We did not find significant improvement in keypoint detection.