Research On Image Segmentation Method Based On Fuzzy Cluster And Retinex Theory

Image segmentation is the process of dividing an image into different regions with similar properties and extracting region of interest.It plays an important role in the fields of computer vision and pattern recognition.However,in the actual process of image processing,intensity inhomogeneity caused by imaging equipment,lighting conditions and complex backgrounds results in poor image quality,and brings great challenge to image segmentation.In addition,due to the need of multi-object recognition,it is also valuable to study the multi-phase image segmentation.This thesis mainly combines variational method,fuzzy clustering and Retinex theory to study image segmentation models and algorithms.The main research contents and innovations can be summarized as follows:(1)A segmentation model based on fuzzy cluster and Retinex theory is proposed for segmenting images with intensity inhomogeneity.By introducing fuzzy membership functions that allow each pixel to belong to multiple regions simultaneously with different membership degrees,the variational model can retain more information from the original images.What’s more,the solution existence of the proposed model is proved theoretically and a fast algorithm is designed to make numerical solution under the framework of alternating direction method of multipliers.Finally,numerical experimental results indicate that the proposed model can get precise segmentation results for natural images and real images with intensity inhomogeneity,and obtain relatively complete contour curves.(2)Due to the need of recognizing multiple specific objects in complex images,the two-phase segmentation method cannot effectively deal with this problem.So,the proposed two-phase image segmentation model is extended to multi-phase image segmentation,and the solution existence of the multiphase model is proved.(3)In the multiphase image segmentation model,the fuzzy membership function is restricted by two constraints,which makes it difficult to solve the model.So,an effective composite solution algorithm is designed,that is,alternating direction multiplier algorithm nested forward and backward splitting algorithm.And,each subproblem can be solved accurately through internal and external iteration to convergence.Furthermore,the proposed model and method are feasible and effective on numerical simulation experiments of brain MR images and real images.