From Perception to Cognition: A Review of Neural- Symbolic AI for Computer Vision

Abstract

Deep learning has improved core computer vision tasks a great deal, but most perception-based methods still have weak symbol grounding, limited compositional reasoning, poor generalization, and low interpretability. Neural-Symbolic AI combines neural feature learning with symbolic reasoning and has become an active approach for dealing with these problems. This paper reviews recent work on vision-focused Neural-Symbolic AI through the perception-reasoning-cognition pathway. It covers representative architectures, including scene-graph reasoning models, neural modules with differentiable logic, knowledge-graph-enhanced frameworks, and 3D neuro-symbolic grounding methods, along with their applications in visual reasoning and scene understanding. The review examines the main achievements, current limitations, and future prospects of the field. Overall, Neural-Symbolic AI shifts computer vision from pure perception toward structured understanding and reasoning, although problems remain in scalable reasoning, knowledge integration, and real-world deployment. It offers a practical direction for building more explainable and reliable vision systems.