Research On Image Segmentation Based On Partial Differential Equation

Image segmentation is to divide the image into several specific,unique regions.Is an important technology of digital image processing,it is also a foundation and the key process of image analysis,has been widespread attention for years,and become a hot issue in the field of image processing.To the last century 90's,the field of image processing based on partial differential equations appeared,the basic idea is to translate the actual research question into a functional minimum problem,and then use the variation method to obtain a partial differential equation,coupled with the initial conditions,through the numerical method to find the solution.Because of its mature theoretical basis and unique numerical method,become a very notable emerging field.Chan-Vese(C-V)model is a typical active contour model in the field of image segmentation of partial differential equation,which uses the global information of the image,and has very strong robustness to the edge of the fuzzy goals and local noise of the image.However,the model for non-homogenous objects can not usually have a good segmentation.In this paper,we carried on the thorough research for C-V model.On this basis,for the characteristics of infrared ship target images and hyperspectral remote sensing images,we put forward two kinds of adaptive segmentation models.The main innovation points in the following aspects:1.Put forward a kind of adaptive active contour model for infrared(IR)images segmentation.For the low contrast between the target and background,edge blur and other features of IR images,by constructing an adaptive gradient modulus value weighting function,effectively combine the region information and boundary information of images,thus based on the current regional gradient modulus value to adaptive processing,ensures the model can segmentation IR image with heterogeneous area effectively.The proposed model is effective to overcome the sensitive to noise of traditional C-V model and low contrast image effect is poor,at the same time with heterogeneous area of infrared ship image target is to keep the higher degree of precision segmentation.2.A active contour model of hyperspectral image of coastal zone land segmentation be proposed.Because the hyperspectral remote sensing image has the characteristics of high spectral resolution,complex band,and large amount of image data,by introducing a kind of edge guide function based on gradient modulus value,improve the model's ability to capture the edge area;In addition,with the introduction of an control function area,which is adapt to the image gray level difference inside and outside to effectively improve the evolution speedof the model.The proposed model achieves the effective segmentation of hyperspectral image with coastal zone land and water area,at the same time overcomes the slow convergence speed and the shortage of segmentation accuracy of traditional C-V model for hyperspectral image segmentation.