Probability Models for Open Set Recognition

Real-world tasks in computer vision often touch upon open set recognition: multi-class recognition with incomplete knowledge of the world and many unknown inputs. Recent work on this problem has proposed a model incorporating an open space risk term to account for the space beyond the reasonable support of known classes. This paper extends the general idea of open space risk limiting classification to accommodate non-linear classifiers in a multi-class setting. We introduce a new open set recognition model called compact abating probability (CAP), where the probability of class membership decreases in value (abates) as points move from known data toward open space. We show that CAP models improve open set recognition for multiple algorithms. Leveraging the CAP formulation, we go on to describe the novel Weibull-calibrated SVM (W-SVM) algorithm, which combines the useful properties of statistical extreme value theory for score calibration with one-class and binary support vector machines. Our experiments show that the W-SVM is significantly better for open set object detection and OCR problems when compared to the state-of-the-art for the same tasks.

1 Introduction

In a recent paper in this journal [27], we raised the issue of open set recognition for visual learning, where not all classes encountered during testing are known during training. This is a necessary and difficult problem to tackle. As an initial solution, we proposed an algorithm called the 1-vs-Set machine, which is suitable for single-class detection tasks in an open set scenario. In essence, the 1-vs-Set machine manages the risk of the unknown by solving a two-plane optimization problem that yields a linear classifier. Detection is a useful operation (almost every digital camera has an automatic face detector these days), but in many cases, we would like to recognize which known classes, if any, are associated with the input image. This can enable applications such as unconstrained optical character recognition (OCR), and photo or video tagging without constraints on the input. In this paper we consider the multi-class open set recognition problem.