The Image Segmentation Algorithm Based On Markov Random Field And Its Application In Change Detection Research

Image segmentation is the key step from image processing to image analysis.The technique is one of the important technologies in computer vision. Image segmentation based on Markov Random Field is a method based on statistical theory, which concerns the correlation between the local pixels, uses the priori knowledge effectively, has fewer model parameters and is easy to be combined with other methods etc., so this method has been extensive researched and applicated in the field of image segmentation.In this dissertation, we studied the MRF-based image segmentation algorithm, especially on MRF method based on multiscale wavelet theory, and apply the method for change detection take into account the regional change ratio using two remote sensing images.Firstly, we studied the multiscale modeling algorithm based on wavelet transformation. Then we gave the variable weights to the tag field model and the characteristic field model between each scale based on multi-scale wavelet modeling. Differnet image scale expressed unique information different from others. Experiment results show that it can get better segmentationas results to set the variable weights adapt to the tag field model and the characteristic field model in different scales. The accuracy of regional info and edge localization is improved using the proposed method.Secondly, we applied the above method for change detection considering the regional statistics. There are many steps such as image registration, radiometric correction, shadow irresistible factors, etc., lead to high fase detection rate. In the dissertation, the rate of change info in each divided region can be obtained through the statistical analysis. Experiments show that the proposed statistical method can reduce the false detection reate and improve the accuracy of change detection.