Mathematical Morphological Segmentation of IKONOS Multispectral Data
基于数学形态学的IKONOS多光谱图像分割方法研究

Image segmentation has been an mi portant research area in mi age analysis and interpretation. An ideal seg- mentation strategy of remotely sensed data should considerproblems ofover-segmentation and under-segmentation smi ulta- neously and find a good tradeoffbetween them. In thispaper, mi age segmentation for IKONOSmultispectraldata is inves- tigated by using techniques ofmathematicalmorphology, and a novelhybrid segmentation algorithm is proposed by combi- ning both edge and texture featuresof mi ages. Based on theK-L transform ofmultispectraldata, edge features are detected bymorphologicalmultiscale andmultidirection gradient algorithms, and mi age objects aremarked throughmorphological filtering and localvariance features extracting. Finally, themarkercontrolledwatershed algorithm is mi plemented. The re- sults indicate that the performance of the proposed algorithm is superior to the gradientbasedwatershed segmentation. Mo- reover, this approach ismore suitable forhigh resolution remotely sensed data to overcome over-segmentation and under- segmentation problems effectively.