Hyperspectral Image Segmentation Based on Spatial-spectral Constrained Region Active Contour

Image segmentation is an important technique in remote sensing hyperspectral image (HSI) processing and application field. It not only can lay a foundation for subsequent object detection, positioning and recognition, but also play an assistant role on HSI data compression and transmission. Level Set which is an effective tool in dealing with image segmentation, has been developed and drawn more attention. For instance, Keaton and Brokish apply spectral and texture information to define the velocity function of level set equation and extract the road from multispectral image [1]. Ball and Bruce extend classification mask design and segment two urban and rural HSIs by using spectral similarity [2]. And some methods use the local edge information in curve evolving by velocity function restriction [3]. However, the lower spatial resolution of HSI usually leads to the vague of object margin and the existence of mixing pixels. The above methods will be less satisfied when no clear boundary of different ground covers exists.