Research On Image Segmentation Based On The Hybrid Bias Field Correction

Image segmentation is a classical problem in image processing,and it plays a key role in image analysis,recognition,and understanding.Due to limitations of imaging equipment and environmental factors during the acquisition process,grayscale inhomogeneity may occur in images,which can increase the difficulty of segmentation and lead to incorrect results.How to effectively segment an image with intensity inhomogeneity into several meaningful images from the perspective of human visual perception is still a very challenging task,as it requires ensuring consistency of segmented regions at different resolutions.The thesis proposes a hybrid bias field correction-based image segmentation model for segmenting natural and medical images with intensity inhomogeneous.This thesis analyzes and studies existing segmentation models for intensity inhomogeneity.Based on the assump-tions of image decomposition models,combining the multiplicative and additive bias fields to form a hybrid field correction model,while using the Sobolev space(2^1,2（?）to describe them,and using a constraint on the multiplicative bias field.By considering hybrid bias fields,our proposed model can effectively remove existing bias fields in images and accurately segment natural and medical images with inhomogeneous intensity.This approach is robust for different image segmentation tasks.Additionally,this thesis proves the existence of a solution for our proposed model.Then we design an algorithm to solve the model quickly and use the Alternat-ing Direction Multiplier Method to iteratively solve the subproblem.And the related properties of the algorithm are analyzed.This thesis conducted related experiments on natural images and real Magnetic Resonance Imaging（MRI）images.Compared with seven classical segmentation models,our method achieved a high segmentation index.To verify the stability of the model,different initial con-tours and images with noise are selected for experiments in this thesis,and good segmentation results are obtained.This demonstrates that our model can effectively remove bias fields in the image and solve the problem of uneven gray level image segmentation.