Active Contour-based Segmentation Methods For Infrared Images

Image processing has important applications in military, remote sensing,meteorology, medicine and many other fields. Image segmentation is an important partof image processing. The goal of image segmentation is to partition a given image into acollection of “objects”, upon which other high-level tasks such as object detection,recognition, and tracking can be further performed. Since the1990s, many scholarshave focus on the image segmentation technique based on active contour models whichare based on partial differential equations and this technique has proven its worth inpractice.Currently, the infrared technology has been widely used in guidance, automaticcontrol, artificial intelligence and many other fields. The infrared image segmentationplays a pivotal role in infrared target identification and tracking. At present, methodssuch as threshold method and genetic algorithm are commonly used in the infraredimage segmentation. Active contour models have their own advantages, so they havebeen applied by some scholars to segment infrared image, and made some achievements.But in general, the methods based on active contour models in the application ofinfrared image segmentation are still placed in the exploring stage.This dissertation focuses on the infrared image segmentation method based onactive contour models; the main results are summarized as follows:The edge-based active contour model is one kind of geometric active contourmodels. The edge stopping function plays an important role in the design of theedge-based model. The traditional edge stopping function only uses the gradientinformation of image pixels to control the evolution of the curve. And it can achievegood segmentation results when the given image has small noise and clear edge.However, the infrared image is often with large noise, blurry edges and low contrast. Itis usually difficult to obtain good segmentation results of infrared image using thetraditional edge stopping function. The local entropy of image can well reflects the graylevel change in a small window of the image and suppress noise easily. In this paper, anew edge stopping function combining with the local entropy of image is proposed. Andit is applied to the edge-based active contour model to segment infrared image. Theexperimental results show that the edge-based active contour based on the new edgestopping function can well segment infrared image. Besides, an automatic initialization scheme based on the feature of infrared image is also presented, which makes theproposed algorithm more efficient.