Modular Facial Expression Recognition on Noisy Data Using Robust PCA
Saloni Mundra, Sujata, Suman Kumar Mitra · 2019
This paper proposes an efficient facial expression recognition approach based on Robust Principal Component Analysis (RPCA). Facial images are susceptible to image noise and varying illuminations. Standard Principal Component Analysis (PCA) when used on such data alters the quality of the approximation. To handle this, we use RPCA to first remove the noise from the images and then transform it to a lower subspace. To reduce the computational complexity, we have used the modular approach in which the regions of the eyes, nose, mouth and forehead are automatically detected and cropped out. RPCA is used on these regions and 1-NN classifier is used for recognition. Experiments performed on CK+, JAFFE and Oulu CASIA datasets show that our method improves the recognition rate over standard PCA when noise is present.