Research On Image Segmentation Method Based On Local Prior Information

Image segmentation is to segment an image into consistent areas according to its intensity,color and texture characteristics and extract the target area of interest for further image processing.In the research and applications of image processing,image segmentation is an essential step,and the quality of segmentation directly affects the effect of subsequent image processing and analysis.However,due to the complexity of imaging scenes,images are often affected by intensity inhomogeneity and noise of different degrees,which greatly increases the difficulty and challenge of image segmentation.In addition,multiphase image segmentation is also a current research hotspot.This paper mainly studies image segmentation methods based on partial differential equation and image segmentation models,and algorithms based on local prior information.The main research contents and innovations are as follows:(1)A two-stage model is proposed and a fast algorithm of the model is designed for segmenting images with intensity inhomogeneity.In the first stage,an image variational method based on Retinex theory is proposed to extract the structure part of the image,which employing the smoothness of the structure part and the bias part.In the second stage,a total variation image segmentation model is proposed to segment the extracted structural parts,which utilizing the local constant characteristics of bias field.Numerical experimental results show the proposed model have a good effect on segmenting images with intensity inhomogeneity.(2)A new variational model based on local prior information is proposed to segment images with intensity inhomogeneity and noise generated by imaging equipment.The proposed model applies simultaneously the local constant and global smoothness priors to describe the bias part such that our model is robust to noise and can achieve more precise segmentation results.Numerical experimental results demonstrate that our method can segment images with intensity inhomogeneity and noise well.(3)The proposed model is extended to multiphase image segmentation by introducing soft membership functions.The existence of minimizers of the proposed variational model is proved,and a fast solution algorithm for the model is designed.Moreover,the time complexity and convergence of the algorithm are displayed.The final experimental results confirm that the model can obtain accurate segmentation results for multiphase brain MR images,and the proposed model is also comparable with the existing models in CPU calculation time.