Research On Digital Image Segmentation Based On Geometric Active Contour Model

Image segmentation technology aims to divide the image into non-overlapping sub-regions with different character each other,and extract the parts of interest,which is the premise and basis for further analysis of the image.With the development of image segmentation technology,thousands of different types of segmentation methods have been proposed in the past decades.It is impossible to form a common image segmentation algorithm due to the wide variety of images,the different purposes of segmentation,and the differences in regions of interest to be extracted.In general,researchers develop image segmentation methods that solve concrete problems for specific images.Among them,the geometric active contour segmentation method based on level set has become a hotspot of image segmentation algorithm for its unique advantages.Many well-known scholars have done a lot of research work in this direction,but there are still some problems need to be solved.For example,the selection of the initial contour has a certain influence on the segmentation results.The segmentation algorithm cannot segment the non-homogeneous image well(referring to the image with uneven distribution of gray value),and segmentation models are sensitive to noise.The resolution of these problems needs to be further studied.Based on the above situation,this academic paper carried out the following work:1.A new adaptive geometric active contour segmentation model based on fractional differential is proposed.In order to improve the anti-noise ability of the classical segmentation model,this dissertation combines fractional differential,level set and curve evolution to construct a new model.The model contains three parts: global term,local term and regular term.The global term is an existing pressure symbol function model.The local term combines fractional differential,fractional differential gradient mode,and difference image to achieve effective segmentation of noise images.The regular term utilizes the length characteristics of the curve to ensure the smoothness of the evolution curve.The fractional differentiation of the image preserves the low-frequency information such as texture details in the image,and also enhances the high-frequency information such as the boundary.At the same time,a new adaptive weighting strategy is proposed by using the local standard deviation of the image,which effectively adjusts the proportion of global terms and local items in the model.In addition,the differences of three commonly used fractional differential definitions are also studied and applied to the proposed new model.The experimental results show that there is no significant difference in the segmentation results of different defined fractional differential models.Compared with the traditional geometric active contour segmentation model,the proposed new model has good anti-noise performance and good segmentation effect on non-homogeneous images.2.A new geometric active contour segmentation model based on Legendre polynomial and image global information is constructed.The assumption of the slice constant image makes it difficult for the traditional segmentation model to accurately segment non-homogeneous images.In response to this problem,Legendre polynomial function is applied to replace the slice constants of the classical Chan-Vese model.The Legendre polynomial is a set of commonly used standard orthogonal bases with smoothness.The smooth function formed by the linear combination of the set of bases can accurately describe the variation of image features.However,there are still some unsolved problems in the existing segmentation model based on Legendre polynomial,such as low efficiency and the selection of segmentation results depending on the initial contour.A new segmentation model is built by combining the global fitting term of the Chan-Vese model to Legendre level set model.Experiment results show that segmentation results of the new model are more accurate and not affected by the selection of the initial contour than the existing Legendre level set model and traditional segmentation models.