Research And Application Of Image Segmentation Method With Constraints

Image segmentation is the job that identifies and extracts the boundaries of the object.As an important interface from image processing to image analysis,the results it provides will directly affect the feature analysis and understanding of the target image.However,real images often exhibit such problems as large noise,blurred boundaries,and uneven image density,which brings many difficulties to boundary recognition and image segmentation.Scholars have made a lot of meticulous research on the issue of image segmentation,and have also achieved many excellent results.Among them,The segmentation method based on fuzzy clustering introduces the fuzzy theory into the traditional clustering method,which not only retains the advantages of the traditional clustering method such as simple algorithm and easy implementation,but also overcomes the "non-or-there" in the classification principle.It makes more original image information preserved,so it has always been a research hotspot in the field of image segmentation.The level set segmentation method,which combines the variational level set theory with the active contours,fuses the underlying image information and high-level understanding mechanisms such as prior information into a unified mathematical framework.It breaks through the defect that the topological structure cannot be changed in the traditional parametric active contour model.The result it provides is unique and stable.At the same time,it also has mature mathematics theory support and simple mathematical implementation.Therefore,it has received extensive attention in many fields such as image denoising and image segmentation.However,the existing two kinds of level set models have problems,such as poor ability to keep weak borders,unevenness to noise and gray scale,sensitivity to initialization,and poor retention of sharp angle information.For complex images,especially very unclear target area such as a medical image.it is difficult to perform accurate segmentation.In order to solve this problem,we have thoroughly studied the image segmentation algorithm with constraint conditions,discussed the constraints and their impact on the segmentation results,and obtained the following innovations and results:1.To solve the problems existing in traditional fuzzy C-means algorithms,such as the inability to fully understand the semantics and content of a given image,ignoring spatial neighborhood information,and low efficiency,an interactive constraint fuzzy clustering algorithm is proposed.The algorithm allows users to guide image segmentation by drawing strokes.It also considers color information,neighborhood information,and intuitive information provided by interactive strokes.Experiments on natural images show that the algorithm can meet the requirements of users and can extract target objects from a given image.2.Aiming at the problem of poor image segmentation due to noise interference,fuzzy boundaries and uneven gray levels in traditional level set models,two constraint-based level set partitioning models are proposed.In these two kinds of models,regional consistency constraint items are defined firstly,which measure the regional consistency on both sides of the target boundary by detecting the zero level set neighboring area.In the two models,the boundary contour is defined as a strip zone in the constraint term,which greatly improves the stability of the boundary and greatly improves the recognition of the weak boundary.It overcomes the problems of poor edge retention and sensitivity to initialization in traditional edge-based level set models.it also overcomes the strong noise-sensitive deficiencies.The two models can take larger smoothing and expansion coefficients to improve the segmentation efficiency.At the same time,the fuzzy boundary caused by the smoothing effect can be well maintained under the effect of the regional consistency constraint.Experiments on liver,CT,Experiments on Liver CT and blood vessels also demonstrated the excellent properties of the two models.3.For the sharp boundary part of the image contour,the smoothness of the level set function tends to cause the sharp angle information to be lost.We propose a new level set evolution method based on the local homogeneity test.Firstly,the local homogeneity test fitting is constructed,and image area information and edge information is merged in a unified mathematical framework.It overcomes the problem that the smooth level set function can not maintain the sharp features of the boundary in the image segmentation,and can also obtain a good segmentation effect in terms of maintaining the weak boundary,anti-noise,and processing unevenness in gray levels.In summary,the research work of this paper is mainly to impose the interactive constraints on the fuzzy clustering model and regional gray consistency constraints on the variational level set model to segment the real image with complex features and medical images with noise interference and blurred boundaries.The proposed interaction and segmentation constraints provide a useful reference for the segmentation of complex images.