Change detection in multi-temporal SAR images

Change detection in multi-temporal SAR imagery it’s a useful tool for preventing natural disasters, studies of the terrain and many other utilities but it’s a tool in constant improvement always looking for better results and less computational costs. Though there are many ways to deal with it, the present thesis focuses in change detection using a generalization of the Kittler & Illingworth threshold (GKIT) algorithm which aim is to obtain an unsupervised method for change detection in multi-temporal images. However, the kittler & Illingworth method on his own it’s not enough in most of the cases for precise unsupervised change detection. Many factors affect the results. The most significant it’s the “speckle” noise, a common multiplicative noise in SAR images. K&I uses a ratio approach to discriminate the “speckle” noise and make it easier to eliminate. As the K&I it’s not enough to eliminate the noise many filters are implemented based on wavelet filtering. With two wavelet filtering approaches, based on the discrete wavelet transform (DWT) and on the stationary wavelet transform (SWT), the results are better but still imprecise. The solution to this drawback it’s a multi-channel fusion method based on the Markov random fields (MRF). The MRF method fuses filtered ratio images and the original one in order to obtain a final ratio image that eliminates most of the noise retaining most of the details. The use of all this concepts will make a very efficient and precise change detection in multi-temporal SAR images.