Remote Sensing Image Classification Based on Markov Random Field

With the continuous development of optical remote sensing technology, the segmentation of high resolution remote sensing images has become a hot research field. The high resolution and complex remote sensing images make the general natural image segmentation become a challenge task. Here, Markov random field (MRF) can well combine the spatial information of remote sensing image with the domain information between pixels for image segmentation, which is currently a hot research direction. In this paper, the wavelet and MRF are combined to achieve multi-scale analysis and produce more accurate segmentation results, the extensive experiments demonstrated that proposed method is more suitable for remote sensing image segmentation, and it provides a good boundary local mapping results and is robust with image non-stationary signal. In addition, in view of the disadvantage of fixed potential function parameters of traditional MRF, we put forward the method of variable weight. On this basis, Kullback-Leibler (KL) divergence is proposed to calculate the similarity between the segmented regions to further optimize the segmentation results.