Spatial location constraint prototype loss for open set recognition

• Published in: Computer vision and image understanding 2023-03, Vol.229, p.103651, Article 103651
• Main Authors: Xia, Ziheng, Wang, Penghui, Dong, Ganggang, Liu, Hongwei
• Format: Article
• Language: English
• Subjects: Empirical risk; Matching theory; Open set recognition; Open space risk; Spatial location constrain prototype loss
• ISSN: 1077-3142; 1090-235X
• DOI: 10.1016/j.cviu.2023.103651

One of the challenges in pattern recognition is open set recognition. Compared with closed set recognition, open set recognition needs to reduce not only the empirical risk, but also the open space risk, and how to reduce the open space risk is the key of open set recognition. Compared with the previous work, this paper analyzes the distribution rules of the known and unknown features, which is highly related to the open space risk. Then, this paper proposes the matching theory to explain the origin of the open space risk. On this basis, the spatial location constraint prototype loss function is proposed to reduce both risks simultaneously. Extensive experiments on multiple benchmark datasets and many visualization results verify the validity of the proposed matching theory and the effectiveness of the proposed method. •The distribution of the known and unknown features is analyzed in detail.•A matching theory is proposed to explain the origin of the open space risk when a deep neural network is used for OSR.•A novel loss function, SLCPL, is proposed to address the OSR problem.•Many experiments and analyses are performed to verify the proposed theory about the origin of the open space risk and the effectiveness of the proposed loss function.