Improving image recognition performance using a hexagonal pixel arrangement

The ultimate goal of this research is to provide a high-precision and high-speed image processing system to society. To achieve this goal, this paper aims to overcome the challenges that were previously unavoidable (such as the ambiguity in the definition of pixel interconnectivity) by reconsidering the structure of pixel arrangements, improving both the accuracy and speed of image processing. Specifically, this paper involves replacing the widely used square pixel arrangement with a hexagonal (honeycomb) structure and establishing a logical framework that allows for essential image processing operations, such as convolution performed on the honeycomb structure. Based on the analysis results of the general object recognition algorithm, this paper proposes a system architecture and formulates convolution operations, including stride and padding, necessary for constructing the proposed system. The formulation utilized the 3-axis discrete coordinate system "HCP (HoneyComb by Projection)." A comparison with conventional systems will then demonstrate the advantages of the proposed framework.