Nuclear Norm Based Matrix Regression with Applications to Face Recognition with Occlusion and Illumination Changes

Abstract

Recently, regression analysis has become a popular tool for face recognition. Most existing regression methods use the one-dimensional, pixel-based error model, which characterizes the representation error individually, pixel by pixel, and thus neglects the two-dimensional structure of the error image. We observe that occlusion and illumination changes generally lead, approximately, to a low-rank error image. In order to make use of this low-rank structural information, this paper presents a two-dimensional image-matrix-based error model, namely, nuclear norm based matrix regression (NMR), for face representation and classification. NMR uses the minimal nuclear norm of representation error image as a criterion, and the alternating direction method of multipliers (ADMM) to calculate the regression coefficients. We further develop a fast ADMM algorithm to solve the approximate NMR model and show it has a quadratic rate of convergence. We experiment using five popular face image databases: the Extended Yale B, AR, EURECOM, Multi-PIE and FRGC. Experimental results demonstrate the performance advantage of NMR over the state-of-the-art regression-based methods for face recognition in the presence of occlusion and illumination variations.

Techniques for the Extraction of Vibrational Signature: A New Method of Pottery Shard Identification

“…The extraction of a unique signature for each image facilitates the identification of an individual by matching the signature to a database set of known of signatures (Yang et al, 2017). We previously reported the ability to identify pottery shards via a similar method of signature matching.…”

Robust Face Expression Recognition with OSCAR

“…They are 243 pieces of 256×256 gray-images. These images will be pretreated before feature extracting: 1) Original images are incised, most facial feature region are reserved.…”

“…Except for the huge volumes, another significant difficulty is that most of the visual data are collected without any control. This problem is especially striking for face images [1]. For example, Facebook and Google Picasa web photo albums typically contain thousands of face images, most of which were obtained without control over facial expression, viewing angle, lighting conditions, occlusions, and image quality.…”

“…The choice of constraint used here is one weighted combination of the 1 L norm and a pairwise L ∞ norm for the coefficients. Specifically, the cons-least-squares optimization problem for the OSCAR is given by 2 1 arg min…”

“…, λ λ are regularization parameters. The OSCAR regularizer thus consists of two parts: an 1 L -regularizer which encourages sparsity as in lasso and a pairwise L ∞ -regularizer which encourages every coefficient pair i β , j β to be equal (Fig.2). This can encourage highly correlated features to be grouped together.…”