Research On Image Fusion And Object Contour Tracking Algorithm Based On Variational

This dissertation focuses on algorithms of remote image fusion and object contour tracking of medical sequence images,and proposes novel models based on variational method.The main contributions and innovations of this paper are as follows:(1)With the variational framework,the fusion of panchromatic(PAN)and multispectral(MS)bands of remote sensing images is deeply analyzed.In view of the lack of the balance between spectral quality and spatial quality in the fused images generated by the conventional algorithms,modulation transfer function(MTF)is introduced to restrict the spatial details injected into the MS bands,and three novel pan-sharpening models are proposed based on different methods to merge spatial information.Model 1 introduces the classical GIHS fusion algorithm into the variational framework,and constructs the low-pass filter with MTF to constrain the injected spatial details.Model 2 obtains the weight coefficients of MS bands based on on the linear regression between the degraded PAN and the original MS images,and injects the spatial details extracted from the PAN image into MS images with those weight parameters.The weight function is defined to reserve the geometrics structure of the PAN image effectively with the gradient of the PAN image;The low-pass filter is developed based on MTF of the multispectral band sensors,which restrains the amount of the spatial details merged into MS images adaptively and reduces the spectral distortion of the fused MS images.To deal with the ill-posed problem formulated by the fusion operation,the L1 regularization term is introduced into the variational framework,which ensure the stability of the numerical solution.Split Bregman method,which can improve the computational efficiency,is applied to acquire the optimization solution of the energy functional.Model 3 improves the weight function of the model 2 and combines edge enhancement term and spatial detail term into gradient information term,which simplifies the calculation and increases the practical feasibility of model 2.The experimental results on QuickBird/IKONOS/GeoEye-1 datasets demonstrate that three novel pan-sharpening models can effectively preserve the spectral information of the original image while improving the spatial quality of the image,and the fusion image achieve an ideal balance between spatial quality and spectral quality.(2)In view of the limitations of geometric active contour applied in cardiac MRI images,a variational segmentation model based on the prior shape and Kullback-Leibler(KL)distance is proposed in this paper.By introducing the gray distribution of the image into the variational framework,the epicardial contour of the left ventricle(LV)can be effectively tracked.The initial evolution curve is automatically located by calculationg the posterior probability of the pixel belonging to the chamber of LV.Based on the KL distance,the model can adaptively set the level set evolution speed and inprove the precision of the outer contour segmentation.The Euclidean distance of Heaviside function is used to measure the similarity between the evolution curve and the prior shape,and the robustness of the model is improved with the introduction of the prior shape into the variational framework.The signed distance constraint term eliminates re-initialization procedure of the level set function,and improves the computational efficiency.The narrow band level set method is used to govern the evolution process of the curve,which can effectively reduce the computational complexity of the model.Considering the shape correlation of the LV outer contour in the cardiac image sequences,the segmentation result of the current frame is referred as the prior shape of the next frame,which guarantees the continuous update of the prior shape.Experimental results illustrate the effectiveness of the proposed model.