Abstract:Active appearance models (AAMs) are useful for face tracking for the advantages of detailed face interpretation, accurate alignment and high efficiency. However, they are sensitive to initial parameters and may easily be stuck in local minima due to the gradient-descent optimization, which makes the AAM based face tracker unstable in the presence of large pose deviation and fast motion. In this paper, we propose to combine the view-based AAMs with two novel temporal filters to overcome the limitations. First, we build a new view space based on the shape parameters of AAMs, instead of the model parameters controlling both the shape and appearance, for the purpose of pose estimation. Then the Kalman filter is used to simultaneously update the pose and shape parameters for a better fitting of each frame. Second, we propose a temporal matching filter which is twofold. The inter-frame local appearance constraint is incorporated into AAM fitting, where the mechanism of the active shape model (ASM) is also implemented in a unified framework to find more accurate matching points. Moreover, we propose to initialize the shape with correspondences found by a random forest based local feature matching. By introducing the local information and temporal correspondences, the twofold temporal matching filter improves the tracking stability when confronted with fast appearance changes. Experimental results show that our algorithm is more pose robust than basic AAMs and some state-of-art AAM based methods, and that it can also handle large expressions and non-extreme illumination changes in test video sequences.

Gabor-based Texture Representation in AAMs.

Gabor Texture in Active Appearance Models

Texture Representation in AAM Using Gabor Wavelet and Local Binary Patterns

Active Appearance Models Using Statistical Characteristics of Gabor Based Texture Representation

A Review of Active Appearance Models

Combining Active Shape Models and Active Appearance Models For Accurate Image Interpretation

Computational Models for Fusion of Texture and Color: a Comparative Study.

Automatic 3D Face Modeling Using 2D Active Appearance Models

Automatic Facial Expression Recognition Using SVM Based on AAMs

Pose robust face tracking by combining view-based AAMs and temporal filters

Subspace Analysis And Optimization For Aam Based Face Alignment

Texture Features for Browsing and Retrieval of Image Data

Abf Based Face Texturing

A Novel Automatic Facial Expression Recognition Method Based on AAM

Age Estimation Using Active Appearance Model Combining with Local Texture Features

Robust Facial Feature Points Extraction in Color Images

Face alignment using statistical models and wavelet features

AAM-Based Face Modeling and Landmark Selection

Texture Representation via Analysis and Synthesis with Generative Adversarial Networks

Learning Gabor Texture Features for Fine-Grained Recognition