The paper introduces a novel mathematical theory to efficiently handle rotations and translations in image processing, specifically in template matching. The authors propose a method that reduces computational complexity by integrating the information relative to all rotated versions of the template into a unique symmetric tensor template. This method potentially speeds up traditional template matching computations, especially for 3D images. The paper also highlights the advantages of the tensor method over machine learning-based approaches, such as being a ‘white box’ model, not requiring training, and accurately estimating rotations.
Publication date: 13 Nov 2023
Project Page: https://arxiv.org/abs/2311.07561v1
Paper: https://arxiv.org/pdf/2311.07561
