Research On Adaptive SAR Image Segmentation Based On Multi-thresholding

Synthetic Aperture Radar has the attractive property of producing images in all-time and all-weather conditions, therefore it has play an important role in the construction of national economy and national defense. SAR image segmentation is the basement in the interpretation and understanding of SAR images. The presence of speckle, which may be modeled as a strong multiplicative noise, makes the segmentation of synthetic aperture radar(SAR) images very difficult. However, for the existing of speckle in SAR image, it leads to the degradation of image quality and makes an undesired effect on obtaining of the SAR images' prior knowledge. Therefore, speckle removal is a key and indispensable step in SAR image preprocessing. The Lee filter is often used as a reference because it combines an efficient noise reduction while maintaining the sharpness of the image. Once the filtering stage is performed, the next step is to define an effective segmentation algorithm.In this paper, two new adaptive SAR image segmentation methods were proposed; they are multi-thresholding based perceptual hash algorithm(MPHA) and histogram-based multi-thresholding fuzzy C-means algorithm(HMFCM). The details of the two algorithms are described as follows:(1) MPHA algorithm takes full advantage of SAR image's gray level information and structure features, which improves the accuracy of our segmentation algorithm. There are two phrases in this method: firstly, according to the gray level information of original SAR image, the pixel values are same or similar in the homogeneous region. Therefore, we can adaptively segment images by selecting multiple thresholds, and then, pixels with the same or similar gray level can be partitioned into the same cluster in the set of initial segmented regions. However, there are too many regions more than the real cluster numbers in the set of initial segmented regions. Thus, the process of region merging is needed to generate the final segmentation results. During region merging, the two different regions with largest similarity degree will merge into the same cluster. In the field of image processing, perceptual hash algorithm is commonly used for image matching. In this paper, perceptual hash algorithm is applied for region merging, which improves the efficiency of the whole SAR image segmentation method, and generates an ideal segmentation results. Several experiments are carried out for comparison of some different segmentation methods to analysis the performance of the proposed algorithm and the initial segmentation method.(2) HMFCM algorithm is an unsupervised method for SAR image segmentation. This method improves the traditional fuzzy C-means method and robust to the speckle noise in the original SAR image. The initial centers of clusters and the number can be obtained adaptively by the implement multi-thresholding method. They are treated as the input data of region merging process, which avoids the problem of over-segmentation and obtains the optimal number of clusters. This method reduces the effect of human factors on SAR image segmentation, and improves the accuracy of the whole segmentation method. In the second phrases, FCM method is introduced to generate the final segmentation results. In the last part of this section, we carry out several experiments on synthetic SAR image and real SAR images to verify the validation of the proposed algorithm.