Research Of SAR Image Boundary Delineation Based On PDE

Boundary detection is one of the fundamental issues in digital image processing. Its main purpose is to isolate the target objects or interested areas from the image background. Besides, it's important prerequisite and necessary step for further analysis and understanding of images, thus it's highly valuable in remote sensing, medical, military etc. For decades, people has already taken a great deal of research in this filed. In recent years, methods based on partial differential equation(PDE) have become the focus with a booming development. In this paper, PDE-based contour detection technologies are discussed, especially the application in SAR image processing.The core of PDE-based image segmentation is the curve evolution theory. A variety of active contour models raised on basis of the theory, such as Snake model, GAC model, Chan-Vese model. Such models promote the active contour to shrink towards the boundaries by minimizing their energy functionals. There models all get widely applied nowadays due to the good adaptability for topology changes of target objects. However, SAR image has its own unique features. Intensity inhomogeneity is occasionally visible in SAR images, which could be caused by the difference of incidence angle or physical error of the radar antenna. In practice, the bias field might lead the active contour to deviate from the boundaries. This paper proposes a variational approach, which does not depend on additional parameters and constraints, for bias correction and boundary delineation of SAR images with intensity inhomogeneity. As the process of bias correction might weaken the boundaries, a weight is added to the stopping function of GAC to enhance it at weak contour by combining the gradient information and the Saliency field derived from the tensor voting method. Experimental results and comparative analysis with other methods in Chapter V illustrate the effectiveness of the method.