Object Segmentation Based On Level Set And An ATR System

The Automatic Target Recognition technology under the natural field and cluttered background is the most dominant in the field of automatic robotic control, medical diagnosis device, security supervision system, and kinds of accurate automatic weapon equipment. Of the field, the image recognition system is the most widely used. In the image recognition system, the object segmentation and the feature extraction is the key of the correct recognition, moreover, hardware system of great stability and high performance is the basis of the algorithm implemented effectively, therefore , the object segmentation technology in the cluttered background which is applicable to the ATR system and the implementation of the ATR hardware system is the focus of our research in this dissertation.The segmentation method based on the Level Set, can not only extract the object region, but also extract kinds of the object's geometrical parameters such as the edge's curvature, the perimeter, the area etc very easily. Compared to the former active contour algorithm, the level set algorithm can solve the topological transformation problem during the contour evolution process very well, and it also has the more stable numerical solutions. However, all the current level set algorithms are not suitable for the application cases in my dissertation, the algorithms based on the edges are vulnerable to the disturbance of the cluttered edges and they can not segment accurately when the image transit smoothly; Although the algorithms based on the regions is stable, they have to satisfy the assumption of global smoothness. The proposed local level set algorithm in this dissertation just needs certain prior location of the object to segment the object automatically in the local region, and thus avoids the disturbance of other region's cluttered background to the segmentation. And we introduced the edge feature of the image into our method, the algorithm converges faster when there are obvious edges in the objects of the image. In order to avoid the disturbance of the cluttered edges to the segmentation result, we also make the region's information act as the dominant and controlling role, while the feature of the edge as assistant. The dissertation also explain the implementation of the local level set algorithm in detail.Aiming at the characteristic of the image processing computing in the ATR system, we devise a kind of the image ATR hardware system based on the structure of DSP+FPGA, it satisfies the requirements of real-time and stable system performance. We explain the design of the hardware system in detail, including the hardware system structure, the chip selection, the design of the key circuit, the FPGA logic. The dissertation also explains the task distribution while the local level set algorithm is implementing on the hardware system.