基于马尔可夫随机场的纹理图像并行分割

基于消息传递接口(Message Passing Interface,MPI)和消息传递并行编程模型,提出了一种针对计算机集群(Cluster)的纹理图像并行分割算法。该算法使用马尔可夫随机场作为纹理特征,通过将图像分块,把特征提取的计算量均匀的分布到并行系统中的各个节点上,从而极大地减少了计算时间。在遥感图像上的实验发现,该算法在4机并行的环境下可以取得与单机串行程序一样精确的分割,而耗时仅为串行程序的31.95%。令人满意的实验结果表明该并行算法不但可以有效的应用于纹理图像分割,而且也为使用计算机集群实现高时间复杂度的图像处理提供了有益的启示。

Parallel segmentation of textured images by using Markov random field

This paper presents a parallel segmentation algorithm for textured images on a computer cluster.Adopting the Message Passing Interface(MPI)and the message passing programming model,this algorithm splits an image into a set of rectangular regions and sends each region to a computer,so that the Markov Random Field(MRF)can be estimated as the texture feature of each pixel on all computers simultaneously.As a result,the time cost of segmentation is greatly reduced.The parallel algorithm has been applied to the segmentation of remote sensing images.The comparative experiments show that,on a four-computer cluster,the parallel algorithm is 3.13 times faster than the serial one.The satisfying results demonstrate that proposed algorithm can provide not only faster segmentation but some inspiration for parallel implementation of computational intensive image processing algorithms.