Research On Partial Differential Equation-based Image Segmentation Algorithms

Digital image processing technology turns the continuous image into a discrete digital image that can be processed by computer by sampling,quantization and coding,so that the useful information can be obtained from the image.The image processing discipline system can be expressed in three levels,which are respectively image processing,image analysis and image understanding.Image segmentation is the key step from the first level to the second level,which lays a foundation for image understanding which is the third level of image processing discipline system.Image segmentation detects the regions or targets which are used or interested in by extracting the feature factors of the image,such as gradient,intensity,texture and corner.The main research content of this thesis is partial differential equation(PDE)of image segmentation technology.PDE method,which is based on mathematics theory,and has strong local adaptability and high flexibility.The characteristics of medical image is more complex.Partial differential equations can fuse different factors and characteristics,which are suitable for medical image processing.This thesis presents two models of image segmentation based on partial differential equations,which have achieved ideal results in medical image segmentation.The main duties and contributions of this thesis are summarized as follows:(1)MAC model based on RD-LSE is presented.RD-LSE is a kind of new level set model.That includes two processes,which are respectively response and spread.MAC model establishes its motion equation by analogizing the movement of the level set function to principle of electromagnetic field.This thesis will fuse RD-LSE methods and level set evolution equation of MAC model,which is a response term.And then diffusion term is introduced into the evolution equation.The model of this thesis don't need to reinitialize the level set function,and it can effectively alleviate the edge leakage problems and speed up the evolution of the original MAC model.It is verified that the proposed model can not only ensure the accuracy of segmentation results but also reduce the number of iterations and running time,by comparing with the MAC model on synthetic images and real images,under the same operating platform.(2)SVMLS model based on local entropy is presented.Local entropy has the certain filtering effect.And local entropy has the effect of smoothing noise,so it is not sensitive to noise.The local image intensities are described by Gaussian distributions with different means and variances.It is able to simultaneously segment the images which have inhomogeneous intensity,and estimate the bias field,and the estimated bias field can be used for intensity inhomogeneity correction.But SVMLS model is sensitive to the noise.The local effect of local entropy is not sensitive to noise.Therefore the total energy is obtained by introducing local entropy descriptor as weight into the SVMLS model,which effectively improve the sensitivity to noise.The proposed model is robust to noises and the segmentation results are satisfactory by comparing with the SVMLS model,under the same noises.Under different intensities of noises,the proposed model can also get ideal segmentation results.With the introduction of local entropy,the noise sensitive problem of SVMLS model is effectively improved.