Research Of Sea-Surface Targets Segmentation Algorithm Based On Remote Sensing Optical Images

Sea surveillance bears its own disadvantages comparing with urban surveillance, which makes remote sensing an indispensable approach in such situations. Remotely sensed optical images always contain large data volume for its broad imaging area and fast shooting speed, as a well accepted point of view, it's necessary to employ computers rather than human interpreters to extract the image information. Based on this background, in this paper we utilize image understanding control strategies of machine vision theory in analyzing the automatic target recognition problem; and present an image segmentation algorithm combining "bottom-up" and "top-down" control strategies. The algorithm can serve as satisfactory low level computing for further feature extractions. This thesis is organized as follows:1. Firstly, we analyze the gray-level intensity characteristics of remote sensing sea-surface visible images and extract a texture feature coincidence with it. Then achieve fast, robust target detection from broad scenarios;2. Introduce random field model for an initial segmentation, separate more external information from the segmented image. Extract ship orientation feature after knowledge-based modification;3. Using target orientation as guidance, detect straight lines around target in edge map produced by canny edge detector. Group the end points of these lines to drive active contour model and avoid the human-machine interactive. The snake evolution is carried out on segmented image, which makes the result more accurate;4. Discuss the remaining limitation of this segmentation technology and search for further improvement. During the research process, obtain an overview of computer vision theory's application in remote sensing image.