The Research On Segmentation And Quantitative Comparison Of Maneuvering Target On SAR Image ATR System

At present, the data acquiring techniques of the Synthetic Aperture Radar (SAR) reconnaissance system are ahead of the data processing techniques significantly, the ability to interpret images can not meet the requirements of a large number of image data processing. How to achieve ATR of SAR images is a problem that should be solved urgently. The quantitative comparison of the similarity of SAR images is an important part of applications such as image retrieval, image matching, target recognition and classification, etc., and has very important significance for automatic interpretation of SAR images.This paper changes the quantitative comparison of the similarity of SAR images into the quantitative comparison of the similarity of maneuvering target, shadow and background to achieve quantitative comparison. Aiming at the shortcoming of CFAR and the characteristics of binarization segmentation, the paper presents a fuzzy processing according to image texture before CFAR. The paper invides comparison of ROI into comparison of contour and gray. The paper presents a novel contour comparison method, that means extracting the information of the ROI, matching feature points, seeking the optimum proportion path of sequence between homonymy points, and calculating similarity of contour. The comparison of gray means measuring the similarity of gray by the difference of average gray and the correlation coefficient of gray histogram. The paper describes and contrasts image texture features based on improved Difference Box-Counting (DBC) method, autocorrelation function method and texture features extracting method based on GLCM. In order to reflect the comparison results, the paper uses the difference of box dimension, ASM and IDM as the measure of texture similarity.The classification and recognition of SAR images is an important link to automatic interpretation of SAR images, while the quantitative comparison of the similarity is the core of classification and recognition. The calculated result of the algorithm proposed in this paper accords with the visual judgement result, which can provide reference for the follow-up image processing.