Image captioning is called as the task of automatically generating descriptive text for images, faces significant challenges in accurately capturing both the objects present and their intricate relationships within a scene. The graph based decomposition captures the hierarchical and compositional structure of the visual content, enabling a more detailed and organized representation of the scene. To generate captions, we employ a graph neural network that iteratively processes the scene graph, refining the representations of nodes and edges by considering their contextual interactions. This refined scene graph is then input into a language model to produce coherent and contextually accurate descriptions of the image. We assess caption diversity using metrics like Driven and Self CIDEr. Experimental results demonstrate that our method surpasses existing state-of-the-art techniques in terms of caption quality and coherence, highlighting the effectiveness of integrating anchor based and scene graph based methodologies.