一种多源遥感图像分割的融合新策略

A Novel Fusion Strategy for Segmentation of Multisensor Remote Sensing Images

为了充分利用多传感器遥感图像数据的互补信息来完成一致的图像解译工作,基于区域邻接图构建了马尔可夫场模型(MRF),并在MRF框架内提出了一种面向多源遥感图像分割的融合新策略.针对由美国陆地卫星探测系统专题制图仪获取的一组多光谱图像和合成孔径雷达图像中的分割问题,提出了具体的数据融合策略,即结合多源图像中的局部特征显著性指标和人眼视觉系统中的重要性因子图制定了融合规则,并在分割过程中充分考虑了传感器的可靠性对融合的影响.人工和真实数据集上的比对分析表明,新策略得到的割图区域匀质性最好,区域轮廓最清晰,并且可以有效提高分割精度.

A Markov random field（MRF） model is defined on a region adjacency graph,and a data fusion strategy for the segmentation of multisource remote sensing images in MRF framework is proposed to fully utilize the complementary information from multisensor remote sensing images for more consistent interpretation.A specific scheme for the segmentation of a set of landsat thematic mapper images and a synthetic aperture radar image is presented in detail.The joint segmentation scheme combines the local feature saliency measure of multisource images with the importance map of human visual system to develop a fusion rule for appropriately incorporating the source reliability to weigh the source influence.Comparative analysis on synthetic and real datasets shows that the new strategy can produce regions with the best homogeneity and the clearest boundary,and can effectively increase the segmentation accuracy.