【摘要】 超像素通常在各类计算机视觉任务中被用于提升计算效率与特征提取能力。针对现有的种子式超像素方法网格初始化策略导致边缘泄露的问题，提出了一种基于颜色一致性的种子式超像素分割算法，通过评估像素的周边颜色一致性，结合梯度总和信息，自适应识别初始化种子，实现在提高分割结果对边界依附性的同时减少了低梯度区域的超像素数目。通过在BSDS500与SBD数据集上的量化评估验证了该算法的有效性以及效率，同时减少了边界依附性与规则性之间的折中。更多还原

【Abstract】 Superpixels are commonly employed in various computer vision tasks to improve computational efficiency and feature extraction capabilities to address the issue of boundary leakage caused by the grid-based initialization strategy in existing seed-initiated superpixel methods, a seed-initiated superpixel segmentation algorithm based on color consistency is proposed. By evaluating the color consistency of the surrounding pixels and incorporating gradient sum information, the initial seeds are adaptively identified, thereby enhancing boundary adherence while reducing the number of superpixels in low-gradient regions. Quantitative evaluations on the BSDS500 and SBD datasets verifythe effectiveness and efficiency of the algorithm while also mitigating the trade-off between boundary adherence and regularity.更多还原