Abstract:Abstract Facial expression recognition has seen rapid development in recent years due to its wide range of applications such as human–computer interaction, health care, and social robots. Although significant progress has been made in this field, it is still challenging to recognize facial expressions with occlusions and large head-poses. To address these issues, this paper presents a cascade regression-based face frontalization (CRFF) method, which aims to immediately reconstruct a clean, frontal and expression-aware face given an in-the-wild facial image. In the first stage, a frontal facial shape is predicted by developing a cascade regression model to learn the pairwise spatial relation between non-frontal face-shape and its frontal counterpart. Unlike most existing shape prediction methods that used single-step regression, the cascade model is a multi-step regressor that gradually aligns non-frontal shape to its frontal view. We employ several different regressors and make a ensemble decision to boost prediction performance. For facial texture reconstruction, active appearance model instantiation is employed to warp the input face to the predicted frontal shape and generate a clean face. To remove occlusions, we train this generative model on manually selected clean-face sets, which ensures generating a clean face as output regardless of whether the input face involves occlusions or not. Unlike the existing face reconstruction methods that are computational expensive, the proposed method works in real time, so it is suitable for dynamic analysis of facial expression. The experimental validation shows that the ensembling cascade model has improved frontal shape prediction accuracy for an average of 5% and the proposed method has achieved superior performance on both static and dynamic recognition of facial expressions over the state-of-the-art approaches. The experimental results demonstrate that the proposed method has achieved expression-preserving frontalization, de-occlusion and has improved performance of facial expression recognition.

Cascade Support Vector Regression-Based Facial Expression-Aware Face Frontalization

Cascade Regression-Based Face Frontalization for Dynamic Facial Expression Analysis

Facial Expression-Aware Face Frontalization.

Cascade of Forests for Face Alignment

Expression-preserving face frontalization improves visually assisted speech processing

Face Frontalization Using an Appearance-Flow-Based Convolutional Neural Network

Dual Sparse Constrained Cascade Regression for Robust Face Alignment.

Frontal Face Synthesizing According to Multiple Non-Frontal Inputs and Its Application in Face Recognition.

3D Facial Expression Reconstruction using Cascaded Regression

A Novel Frontal Facial Synthesis Algorithm Based on Individual Residual Face.

Face Frontalization Based on Robustly Fitting a Deformable Shape Model to 3D Landmarks

Constrained Joint Cascade Regression Framework for Simultaneous Facial Action Unit Recognition and Facial Landmark Detection

Robust Virtual Frontal Face Synthesis from a Given Pose Using Regularized Linear Regression.

Adaptive Cascade Regression Model for Robust Face Alignment.

Learning a High Fidelity Identity Representation for Face Frontalization

Cascade EF-GAN: Progressive Facial Expression Editing with Local Focuses

A Stepwise Frontal Face Synthesis Approach for Large Pose Non-frontal Facial Image.

A multi-pose face frontalization method based on encoder-decoder network

Frontal Face Reconstruction Based on Detail Identification, Variable Scale Self-Attention and Flexible Skip Connection

Robust Face Frontalization in Unconstrained Images.

A novel frontal view synthesis method based on neighbor embedding