Research In Non-uniform Medical Image Segmentation Based On Local Binary Fitting Model

Image segmentation is a kind of image processing techniques and process that divides the image into different regions, each of which owns its distinct feature. One of the major research fields of image segmentation is the medical image segmentation, which is the basic and necessary step for human diseases quantitative and qualitative analysis and visualization.However, intensity inhomogeneity occurs in many real images of different modalities. In particular, it is often seen in medical images such as x-ray radiography/tomography and magnetic resonance(MR) images, due to technical limitations or artifacts introduced by the object being imaged. Intensity inhomogeneity, which makes the medical image changed in local statistical characteristics and overlaps of the intensity of different Tissues, has become a major obstacle to automatic segmentation. To overcome the difficulties caused by intensity inhomogeneity, Chunming Li proposed an energy model based on local binary fitting in 2007, which uses the image local information but not the image global information to optimize the image segmentation through minimizing the local energy functional. However, LBF model is sensitive to initial contour curve, due to local property. LBF models often fail to correct the non-uniform object segmentation to the side of the contour curve, when the initial level set curve is away from the object or cross the border. Based on the problems mentioned above, this paper presents an adaptive intensity fitting active contour model, which is able to adaptively adjust the proportion of global intensity information and local intensity information, according to the distance between the local active contour and object boundaries. This method is able to work effectively on angiography image segmentation with intensity inhomogeneity and noise. Moreover, it is robust to initial contour placement without any additional parameter.In recent decade, bias correction of non-uniform magnetic resomance(MR), which based on segmentation, has been extensively studied. Intensity inhomogeneity correction is often a necessary preprocessing step enabling better image segmentation, thus be viewed upon as two intertwined procedures. In segmentation based intensity inhomogeneity correction methods these two procedures are merged so that they benefit from each other. In 2008, Chunming Li propose a variational level set approach to segmentation and bias correction of images with intensity inhomogeneity. Because the coupling evolution of the different level set curves are applied to multiple objective segmentation, the final segmented results may depend on the choice of the initial curves. Meanwhile, this method is sensitive to initial contour curve and Slowly to curve evolution, due to the local characteristics. To overcome the shortcoming in above-mentioned method, we developed a fast segmentation and bias correction method applied to brain MR image with intensity inhomogeneity. Firstly, unlike previous work, a special initial method is used to decoupled the two level set evolution. In this paper, one level set function is initialed to be a constant in order to avoid evolution of the different level set curves at the same time. Secondly, global information is added to energy equation to avoid traping in local minimum, during the segmentation between brain tissue and background. Finally, A disturbance term is used to guide sign change of level set function value in the gray region, according to characters of gray matter with more boundary and smaller area, during the segmentation between white matter and gray matter.