Automated image captioning is a critical task in computer vision and natural language processing, enabling machines to generate meaningful textual descriptions for images. Traditional approaches, including rule-based systems and statistical methods, have struggled with the complexity of visual scene interpretation, often resulting in generic or inaccurate captions. Recent advancements in deep learning, particularly transformer-based models with visual attention mechanisms, have significantly improved caption accuracy and contextual relevance. However, existing transformer-based approaches face challenges such as high computational demands, difficulty in capturing fine-grained visual details, and the need for improved generalization across diverse datasets. This research explores transformer-based visual attention networks to address these limitations, proposing an optimized framework that enhances caption generation through refined attention mechanisms and effective feature selection. Experimental results demonstrate that our approach improves caption quality, achieving higher BLEU and CIDEr scores compared to baseline models. The findings underscore the potential of transformer-based models in advancing automated image captioning for real-world applications, including assistive technologies and content management systems.