A Graph Segmentation Method Based on Compatibility Subgraph Aggregation

Graph-based methods, distinguished by their resilience to noise, have demonstrated efficacy in image segmentation compared to alternative techniques. Despite their notable capabilities, these methods present over- and under segmentation problems. This paper introduces a novel graph-based approach specially designed to solve these issues effectively. Unlike conventional graph-based methodologies, which predominantly favor either local or global features, our approach considers a combination of these two. Initially, the complete graph is partitioned into multiple subgraphs considering the similarities of the local features in the nodes. Subsequently, these subgraphs are evaluated and classified as either significant or irrelevant, based on the number of nodes they contain. To mitigate over segmentation, we merged the irrelevant subgraphs with the significant subgraphs considering their global features. These mechanisms allow our method to maintain integration between the local and global features, leading to a more precise and accurate representation of the objects in the image. The proposed method was evaluated using various complex images and was benchmarked against state-of-the-art segmentation techniques. The results confirm the superiority of our approach in delivering higher-quality and more dependable segmented images than existing methods.