Exploiting structured high-level knowledge for domain-specific visual classification

作者：

Highlights：

• Use computational ontologies to represent visual semantics for specific domains.

• Integrate high-level structured semantic information into machine learning methods.

• Bayesian belief propagation to predict fine-grained classes from visual evidences.

• Combine graph-based marginal likelihood estimation with deep learning models.

摘要

•Use computational ontologies to represent visual semantics for specific domains.•Integrate high-level structured semantic information into machine learning methods.•Bayesian belief propagation to predict fine-grained classes from visual evidences.•Combine graph-based marginal likelihood estimation with deep learning models.

论文关键词：Fine-grained visual classification,Computational ontologies,Belief networks

论文评审过程：Received 22 August 2019, Revised 29 September 2020, Accepted 22 December 2020, Available online 7 January 2021, Version of Record 14 January 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107806