Multiple One-Shots for Utilizing Class Label Information

• DOI: 10.5244/c.23.77
• Publication Date: 2012-02-07T10:33:24Z
• Authors: Yaniv Taigman; Lior Wolf; Tal Hassner
• Keywords: 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

The One-Shot Similarity (OSS) kernel [3, 4] has recently been introduced as a means of boosting the performance face recognition systems. Given two vectors, their score (Fig. 1) reflects likelihood each vector belonging to same class other and not in defined by fixed set “negative” examples. In this paper we explore how may nevertheless benefit from availability such labels. (a) present system utilizing identity pose information improve facial image pair-matching using multiple scores; (b) show separating lead better rates unconstrained, “wild” images; (c) far can get single descriptor with different similarity tests opposed popular approaches; (d) demonstrate learned metrics for improved performance.