Research On Intensity Inhomogeneity Segmentation Model Based On Fourth-order Regularization

Image segmentation is an image processing method,aimed to extract regions of interest from an image.In reality,there are defects in imaging equipment and changes in lighting,etc.,which cause the intensity inhomogeneities of many images.This makes it difficult to segment intensity inhomogeneity.The level set method is a common method for solving models in image processing using partial differential equations,and it is also widely used in the segmentation problem of intensity inhomogeneity.However,the level set function is difficult to maintain the signed distance property during evolution.In order to avoid heavy calculation caused by reinitialization,and maintain stable level set evolution,regularization term is often added to the model.To solve this question,we have done the following work:Firstly,we find that the fourth-order partial differential equations will evolve the image into a piecewise planar image during evolution.By using the fourth-order term as the level set regularization term of the segmentation model to limit the gradient modulus of the level set funct ion to a finite range,and combined with an existing model's segmentation term,the first model —fourth-order term regularization segmentation model is proposed.Then,two numerical schemes for solving the model are designed,one is the finite difference method,another is the AMOS scheme based on the idea of reducing the dependence of the model on parameter values and speeding up the calculation speed of the model.Secondly,because the fourth-order term is easy to over-smooth the level set function during the evolution process,the contour may lose some detailed information around boundary.Considering that the Laplace operator has advantages in the detection of boundary information,a second-order term is added to the first model.The coupling of the second-order term and the fourth-order term is used as the level set regularization term of the segmentation model.The second model—fourth-order regularization segmentation model with second-order terms is obtained.After that,the second model is solved separately by the finite difference scheme and the AMOS scheme.Finally,in the numerical experiment part,compared the model proposed in this paper with the model without regularization term.The experiment proves that the fourth-order term can play the role of level set regularization.At the same time,The first model proposed in the article is compared with the classic RSF model.The experimental results show that the first model can segment images with higher intensity noise.Besides,comparing the second model with the first model,it is found that the second model is superior to detect object boundary information to the first model and has a wider initial contour range than the first model.