A Systematic Review of Tensor analysis models for high-dimensional computer vision tasks: Methods, Architectures, and Future Research DirectionsHigh-dimensional computer vision tasks such as hyperspectral imaging, video understanding, 3D reconstruction, a

Abstract

High-dimensional computer vision tasks such as hyperspectral imaging, video understanding, 3D reconstruction, and multimodal perception demand efficient representations capable of preserving structural correlations across multiple modes. Tensor analysis models have emerged as a powerful mathematical framework to address these challenges by exploiting multi-linear relationships inherent in high-dimensional data. This paper presents a systematic review of tensor-based methods applied to high-dimensional computer vision tasks, focusing on methodological advancements, architectural innovations, and integration with modern deep learning paradigms. The review synthesizes developments from 2018 to 2025, covering tensor decompositions, tensor regression, tensor networks, and hybrid tensor-deep learning architectures. Key findings highlight the growing convergence between tensor algebra and neural architectures, improved efficiency in handling large-scale data, and enhanced interpretability compared to conventional deep learning approaches. The paper also identifies critical limitations, including computational complexity, scalability issues, and challenges in real-time deployment. Contributions of this review include a structured taxonomy of tensor-based models, comparative evaluation across applications, and identification of emerging research directions such as tensorized transformers and AI-assisted tensor optimization frameworks.