Image recognition and classification under geometric transformations is a critical challenge in computer vision. In this study, we propose a novel approach leveraging a swarm optimization technique to minimize the relative error of Charlier orthogonal Moment’s invariants. This method adapts the Charlier localization parameter \(a_1\) dynamically, optimizing it to produce minimal relative error between an original image and its geometrically transformed counterparts, including translations, rotations, and scaling. The resulting Optimized Charlier Invariant Moments are used as robust features for classification tasks. We construct feature vectors from these moments for individual images in a database and classify them using prominent supervised machine learning algorithms. Experimental results reveal that the proposed method achieves high validation and classification accuracies, demonstrating its efficacy in geometric transformation-invariant image recognition. This approach presents a promising avenue for enhancing the robustness of image-based machine learning applications in dynamic environments.
