Research On Target Contour Extraction In Synthetic Aperture Radar Imagery Based On Active Contour Models

Synthetic aperture radar (SAR), compared with the traditional remote sensing methods, has theall-weather and all-time advantage, which makes SAR being widely used in military and civilapplications. Target contour provides valuable shape information that can be used in pose estimation,target recognition, etc., therefore target contour extraction in synthetic aperture radar (SAR) imageryhas drawn lots of attention around the world. Due to the speckle noise inherent in SAR imagery,traditional target contour extraction methods fail for processing SAR images. To solve this problem,this thesis takes active contour model as the basis, with the speckle noise of SAR imagery underconsiderations, studies the methods of extracting target contours in SAR imagery.Firstly, this thesis takes the analysis for the characteristics of SAR image data. The classical SARimage target contour extraction methods are introduced and the experiments are conducted to showthe deficiencies of the classical methods.Secondly, combining the characteristics of SAR image data, the active contour model is modified.Two modified active contour models are proposed. One modification is based on the idea of combingvector field convolution model and region competition model, the other modification is based on theidea of well segmentated image should have a maximum contrast.Thirdly, this thesis studies the initial contour setting for the active contour model, and proposesthe efficient method of setting the initial contour for the SAR image target contour extraction.Finally, based on synthetic and real SAR image data, the proposed methods of this thesis areverified. The results show, the two target contour extraction methods proposed can accurately andquickly extract the target contours in the SAR imagery, and have the apparent better performance thanthe current classical methods. The initial contour setting method proposed can significantly reduce theiteration times of the active contour model, and this method is very efficient.