The Research On Modeling By Morkov Random Field In Wavelet Domain And Application

Image denoising is the foundation of image processing applied research. Because the noise will affect the subsequent image processing, so research on the denoising algorithm has received the widespread attention. Image segmentation is the key of image processing to the image analysis, but to obtain ideal segmentation results is still a difficulty, so it is necessary to research on image segmentation.In many image analysis methods, MRF model is widely used because it can effectively describe the spatial information of image and have a sound theoretical basis. So MRF model attracts more and more attention of researchers. Wavelet transform has the characteristics of directional, non redundancy and multi resolution analysis. It is a new tool to describe the non stationarity of image. Using the characteristics of wavelet, combining the image analysis based on MRF model and wavelet transform, can improve the quality of image analysis.Firstly, using the statistical characteristics of HMM, the wavelet domain HMT model is developed, and then establishment of the wavelet domain HMT model and the parameter estimation are accomplished. In this paper the wavelet domain HMT model is used for TH-1 image denoising. Firstly, the wavelet coefficients of image are modeled, secondly the model parameters are estimated by using the expectation maximization algorithm, then the denoising image is got by using inverse wavelet transformation and the band synthesis. The experimental results show that this method can effectively eliminate the noise and retain the edge and detail information.Secondly, on the basis of the wavelet domain MRMRF model is established, image segmentation is realized by using this model. Because this method uses potential function in the segmentation process, so it is difficult to obtain consistent segmentation results among different scales. It is improved by using variable weights to connect the characteristics field model and random field model. In this method the contribution of each scale corresponding characteristic field energy and random field energy to the total energy label field is different. This algorithm is tested on a C star first rail SAR remote sensing image, then compare the results with the traditional MRF segmentation method and the wavelet domain MRMRF. The results show that this algorithm can obtain a more accurate segmentation result and increase the practicability of the algorithm.