Improved Mixed Double Domain Image Denoising And SAR Image Change Detection Based On Difference Image Combing And Edge Classification

Image denoising is an important subject in the field of image processing domain. It is aim to obtain the clean image form noise polluted image, and preserve the edges, texture and structure information as best as it can. Synthetic aperture radar(SAR) images are widely used in many aspects because of its characters of all-day and all-weather. Change detection of SAR images means to analysis and compare the two images obtained from same region but different time to capture the change information. Change detection of SAR images has an important application in the prediction and assessment of nature disasters, urban planning, vegetation coving and target detection in military. This thesis mainly studied how to deal with nature images in spatial domain and transform domain to remove the noise; how to combine the difference image and edge detection classification result that get better change information at the edge region in the change detection of SAR images. The mainly work as follows:1. Proposed a denoising method based on spatial domain and transform domain for nature image. The method improved the space kernel and range kernel respectively on the basis of the traditional union bilateral filtering. A connect relationship between attenuation factor and sliding window radius was built for space kernel, and a modify term was added for rang kernel. At first, in the spatial domain the improved bilateral filtering is used to guide the input image filtering to get the spatial processing results, and then calculate the difference image before and after filtering and constraint it with the restraint functions. Then, the filtering results were changed to frequency domain by short-time Fourier transform, and the frequency domain coefficients were shrunk and inverse transformation was completed. At last, the result in the spatial domain and frequency domain were combined and as guidance image to guide the noise image iteratively in order to abate ringing effect. Our method has a better performance both in vision and in data compared other methods.2. Proposed a fusion method based on the differences image and edge detection result method of SAR image change detection. First, a preprocessing of the difference image and log ratio difference image were performed. Then, two difference images were merged by using our fusion method, the merged image was classified by spectral clustering to get initial change detection image. The pixels of edge class and non-edge class were obtained respectively by Primal Sketch classification, and the edge class pixels in the original image were clustered by using of K-means algorithm, so the edge change information was captured. Finally, the initial change detection and edge change detection were combined to obtain the final change detection. The experiment results demonstrate that the effective of our method by compare with other methods.