Final Year Project - Shadow Removal of Human Facial Images 10 Sep 2019
In the academic year 2019-2020, which is my final year, I am doing my final year project.
It is both a academic requirement for graduation that posted by the computer science department, and is also a selective topic in order to match students variant interests. In my case, my partner and I are conducting research under professor Chi-Wing FU’s supervision, with the help of his PhD student Xiaowei Hu and Tianyu Wang.
Topic
The topic is now temporarily decided to be shadow removal of human face photos, especially the shadows produced by wearing glasses. However, there were still some spaces for discussion about this topic. One advice brought by Xiaowei was that it is highly possible for our neural network to have the ability of removing general shadows if it could remove the shadows of glasses. No matter what, we are trying to train the network to specifically remove glasses shadows, and we may see whether it has further abilities to deal with general shadows.
Current Works
-
Collected human facial images and generating fake shadows around eyes for network training.
-
Took real shaded photos paired with shadow-free version for network testing.
-
Utilized Dilated Residual Network (DRN) to suppress shadows. We can remove fake shadows perfectly by now, but the performance on real photos is not good enough. We still need to deal with the domain adaptation problem.