Subspace Representations for Robust Face and Facial Expression Recognition

Thumbnail Image


Publication or External Link






Analyzing human faces and modeling their variations have always been of interest to the computer vision community. Face analysis based on 2D intensity images is a challenging problem, complicated by variations in pose, lighting, blur, and non-rigid facial deformations due to facial expressions. Among the different sources of variation, facial expressions are of interest as important channels of non-verbal communication. Facial expression analysis is also affected by changes in view-point and inter-subject variations in performing different expressions. This dissertation makes an attempt to address some of the challenges involved in developing robust algorithms for face and facial expression recognition by exploiting the idea of proper subspace representations for data.

Variations in the visual appearance of an object mostly arise due to changes in illumination and pose. So we first present a video-based sequential algorithm for estimating the face albedo as an illumination-insensitive signature for face recognition. We show that by knowing/estimating the pose of the face at each frame of a sequence, the albedo can be efficiently estimated using a Kalman filter. Then we extend this to the case of unknown pose by simultaneously tracking the pose as well as updating the albedo through an efficient Bayesian inference method performed using a Rao-Blackwellized particle filter.

Since understanding the effects of blur, especially motion blur, is an important problem in unconstrained visual analysis, we then propose a blur-robust recognition algorithm for faces with spatially varying blur. We model a blurred face as a weighted average of geometrically transformed instances of its clean face. We then build a matrix, for each gallery face, whose column space spans the space of all the motion blurred images obtained from the clean face. This matrix representation is then used to define a proper objective function and perform blur-robust face recognition.

To develop robust and generalizable models for expression analysis one needs to break the dependence of the models on the choice of the coordinate frame of the camera. To this end, we build models for expressions on the affine shape-space (Grassmann manifold), as an approximation to the projective shape-space, by using a Riemannian interpretation of deformations that facial expressions cause on different parts of the face. This representation enables us to perform various expression analysis and recognition algorithms without the need for pose normalization as a preprocessing step.

There is a large degree of inter-subject variations in performing various expressions. This poses an important challenge on developing robust facial expression recognition algorithms. To address this challenge, we propose a dictionary-based approach for facial expression analysis by decomposing expressions in terms of action units (AUs). First, we construct an AU-dictionary using domain experts' knowledge of AUs. To incorporate the high-level knowledge regarding expression decomposition and AUs, we then perform structure-preserving sparse coding by imposing two layers of grouping over AU-dictionary atoms as well as over the test image matrix columns. We use the computed sparse code matrix for each expressive face to perform expression decomposition and recognition.

Most of the existing methods for the recognition of faces and expressions consider either the expression-invariant face recognition problem or the identity-independent facial expression recognition problem. We propose joint face and facial expression recognition using a dictionary-based component separation algorithm (DCS). In this approach, the given expressive face is viewed as a superposition of a neutral face component with a facial expression component, which is sparse with respect to the whole image. This assumption leads to a dictionary-based component separation algorithm, which benefits from the idea of sparsity and morphological diversity. The DCS algorithm uses the data-driven dictionaries to decompose an expressive test face into its constituent components. The sparse codes we obtain as a result of this decomposition are then used for joint face and expression recognition.