1) Traditional face recognition using Gabor wavelet based features

Apply multi-channel Gabor wavelets for feature extraction, adopt subspace analysis, sparse representation and tensor analysis thereafter to encode and combine the extracted Gabor features for face recognition.

2) CNN based deep learning for face recognition

Study state of the art CNN networks like DeepID, VGG etc. based face recognition system, explore different distance measures, metric learning techniques to improve their performance and apply them to real applicatios like ID verification and large-scale person identification.

3) Face expression recognition and synthesis

Test both traditional and CNN based approaches for facial expression recognition, apply auto encoder/decoder network to encode the importation information during expression and attribute variation. Apply them for computer interface.