Facial Expression Recognition using Geometric Normalization and Dual-Tree Complex Wavelet Transform

Due to its various applications, facial expression recognition has been attracted much attention in machine vision research in recent years. Till now, facial expression recognition with high accuracy remains as a challenging problem due to the variability of facial expressions. In this paper, a novel approach is presented, which utilizes an accurate feature extraction method along with the elimination of geometric variability. For this purpose, the mean geometric model is used for normalization and elimination of geometric variability in facial images. Then, Dual-Tree Complex Wavelet Transform is used for appearance feature extraction. It is an approximately shift-invariant and directional transform which can detect edges in different angels. Utilizing the geometric normalization and the mentioned wavelet transform together improves the accuracy more than 5% respects to utilizing each method separately. From the experiments, average recognition rate on CK+ dataset is 93.78% being a noticeable result compared to the existing studies on this dataset.