Image Segmentation Refinement Based on Region Expansion and Minor Contour Adjustments

In high-precision image segmentation tasks, even slight deviations in the segmentation results can bring about significant consequences, especially in certain application areas such as medical imaging and remote sensing image classification. The precision of segmentation has become the main factor limiting its development. Researchers typically refine image segmentation algorithms to enhance accuracy, but it is challenging for any improvement strategy to be effectively applied to images of different objects and scenes. To address this issue, we propose a two-step refinement method for image segmentation, comprising region expansion and minor contour adjustments. First, we design an adaptive gradient thresholding module to provide gradient-based constraints for the refinement process. Next, the region expansion module iteratively refines each segmented region based on colour differences and gradient thresholds. Finally, the minor contour adjustments module leverages local strong gradient features to refine the contour positions further. This method integrates region-level and pixel-level information to refine various image segmentation results. This method was applied to the BSDS500, Cells, and WHU Building datasets. The results demonstrate that the refined closed contours align more closely with the ground truth, with the most notable improvement observed at contour inflection points (corner points). Among the results, the Cells dataset showed the most significant improvement in segmentation accuracy, with the F-score increasing from 87.51% to 89.73% and IoU from 86.83% to 88.40%.