Learning structured ordinal measures for video based face recognition. (March 2018)

• Record Type: Journal Article
• Authors: He, Ran; Tan, Tieniu; Davis, Larry; Sun, Zhenan
• Is Part Of: Pattern recognition. Volume 75(2018:Mar.)
• Page Start: 4
• Page End: 14
• Publication Date: 2018-03
• Subjects: Ordinal measure -- Metric learning -- Local feature; 00-01 -- 99-00; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.02.005
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We proposed a structural ordinal measure (SOM) method by using output structures. SOM encourages ordinal features from the same class to have similar binary codes. We propose a self-correcting method to discretely binarize image-set samples. SOM achieved state-of-the-art results on several datasets. Abstract: Handcrafted ordinal measures (OM) have been widely used in many computer vision problems. This paper presents a structured OM (SOM) method in a data driven way. SOM simultaneously learns ordinal filters and structured ordinal features. It leads to a structural distance metric for video-based face recognition. The SOM problem is posed as a non-convex integer program problem that includes two parts. The first part learns stable ordinal filters to project video data into a large-margin ordinal space. The second seeks self-correcting and discrete codes by balancing the projected data and a rank-one ordinal matrix in a structured low-rank way. Weakly-supervised and supervised structures are considered for the ordinal matrix. In addition, as a complement to hierarchical structures, deep feature representations are integrated into our method to enhance coding stability. An alternating minimization method is employed to handle the discrete and low-rank constraints, yielding high-quality codes that capture prior structures well. Experimental results on three commonly used face video databases show that our SOM method with a simple voting classifier can achieve state-of-the-art recognition rates using fewer features and samples.