Robust face recognition via sparse representation:(基于稀疏表示的人脸识别)
MANUSCRIPT ACCEPTED BY IEEE TRANS. PAMI, MARCH 2008. 1
Robust Face Recognition via Sparse Representation
John Wright, Student Member, Allen Y. Yang, Member, Arvind Ganesh, Student Member,
S. Shankar Sastry, Fellow, and Yi Ma, Senior Member.
Abstract— We consider the problem of automatically recog- parsimony as a principle for choosing a limited subset of features
nizing human faces from frontal views with varying expression or models from the training data, rather than directly using the
and illumination, as well as occlusion and disguise. We cast data for representing or classifying an input (test) signal.
the recognition problem as one of classifying among multiple
linear regression models, and argue that new theory from sparse The role of parsimony in human perception has also been
signal representation offers the key to addressing this problem. strongly supported by studies of human vision. Investigators have
Based on a sparse representation computed by -minimization, recently revealed that in both low-level and mid-level human
we propose a general classification algorithm for (image-based) vision [7], [8], many neurons in the visual pathway are selective
object recognition. This new framework provides new insights for a variety of specific stimuli, such as color, texture, orientation,
into two crucial issues in face recognition: feature extraction and
robustness to occlusion. For feature extraction, we show that if scale, and even view-tuned object images. Considering these neu-
sparsity in