一种基于多级队列的并行区域生长算法

区域生长是经典的图像分割方法之一,为了满足图像分割的实时性要求,提出了一种基于多级队列的并行区域生长算法。该算法采用多级队列存放待生长的种子像素,优先生长边界种子像素,以尽快生成越界种子节点,从而减少邻居节点的等待时间。实验表明,该算法相比一般的基于单队列的算法,加速比有显著提高,且可扩展性较好。

Parallel algorithm of region growing based on multi-level queue

Region growing is one of classical image segmentation methods. In order to satisfy the real-time request of image segmentation, a parallel algorithm of region growing based on multi-level queue was proposed. The algorithm adopted multi-level queue to deposit seed pixels for growing. Boundary seed pixels were grown preferentially, so as the seed pixels beyond boundary were generated early. Therefore, waiting time of neighbors was reduced. Experiments show that compared with general algorithm based on single queue, speed-up ratio of the algorithm can be promoted significantly and its sealability is better.