为了实时而准确地分割高分辨率图像, 提出一种基于图的并行分割算法。该算法采用并行结构, 将图像分为多个区域, 充分利用不相邻区域的处理具有并行性的特点, 对多个区域进行并行处理。算法可直接部署在多核处理器上, 多核并行处理以减少运行时间。为了增强算法对杂乱背景的适应能力, 提升分割的准确性, 采取低权值直接合并策略, 将相对较为均匀的区域直接合并, 使同一目标区域内的分割一致性更强。实验结果表明, 该分割算法的实时性较好, 明显优于现有分割算法, 并且分割准确性较高, 在复杂场景中应用也能取得较好的分割效果。

A novel parallel graph-based image segmentation algorithm was constructed to segment high-resolution image precisely in real time.This algorithm adopts parallel structure and divides the image into multiple regions.Parallel processing can be made to the non-adjacent regions.This algorithm can be deployed easily on multi-core processors for multi-core parallel processing to shorten the runtime.In order to segment precisely and enhance the ability in clutter scenes, this algorithm combines the smooth regions directly by setting low weight threshold.Based on this method, the object regions can be segmented coherently. Experiments demonstrate that this algorithm can segment high-resolution images rapidly and precisely, has better performance than the traditional segmentation algorithms, and can achieve high level performance in segmenting images with complex background.