Method For Medical Image Segmentation Based On Active Contour Model

Image segmentation is a prerequisite for image understanding and recognition, as the foundational part of image processing, it has been a hot and difficult problem of image processing and computer vision. Currently, medical imaging technology with the development and progress of computer technology has been widely developed and applied, computer image segmentation techniques are widely used in medical images, and have played a huge role in pathological analysis, clinical diagnosis, research of organization structure and reproducing three-dimensional human body organs. Due to the factors of complexity, diversity and fuzzy uneven, accurate medical image segmentation becomes one of the most challenging problems in medical image analysis. Traditional image segmentation methods become more and more difficult to meet the requirements of complex heterogeneous image segmentation, which limiting the related fields’better and faster development. Active contour models developed greatly in recent years, they do not rely on image quality when deal with images, with rigorous mathematical foundation, sub-pixel accuracy, and efficient numerical scheme, active contour models have been widely used in various image segmentation problems, it is an effective tool for image segmentation of complex heterogeneous images, and worth further study.Currently, the active contour models are still in the developing stage, the segmentation effect to medical science, nature and remote sensing heterogeneous images is not ideal, there is room for improvement. In the context, we developed the research into the method of image segmentation based on active contour models. By analyzing the main problems existing in active contour models, we present an edge-based global-local fitting active contour model, verify the validity and novelty of the model, experiments show that our model has good effect on medical image segmentation. The main work and contributions of our model are as follows:(1) Combination of the global and local information. Local energy function of LBF model is used in our paper. LBF model only used the local information of image, has obvious advantages in the treatment of low contrast, uneven illumination image, but the detection of non-uniform gray target is prone to error and easy to fall into local minimum when it is used in the strong noise environment, and has higher requirements on the position of the initial contour. Aiming at this problem, we consider joining a global constraint function into the local model, the new model can segment the object correctly from images with low contrast, intensity inhomogeneity and strong noise, and the sensitivity of the initial outline has got a big improvement.(2) The implement of active contour model based on edge information. Now widely used active contour models are most region-based, only using the region information of images, without considering the effect of edge information into the process of segmentation. In this paper, we combine regional information and edge information based into the former model we propose, by introducing an edge detection operator in the model’s energy function, our model can lead the curve evolution converging to a stable state boundary position of the image, ensure the strong edge of the images, so to avoid or reduce the boundary leakage, meanwhile speed up the rate of evolution.(3) Bias correction. In order to ensure stable, effective and quickly evolution of the level set function, we introduce the distance regularized level set evolution (DRLSE) method into our model, this method uses a distance regularization term in the model. With the bias correction between our model and the signed distance function, it ensures the signed distance characteristics in the course of evolution, thus the model don’t have the need of re-initialization, and it has promoted the stability and precision of our model, been more easily to realize comparing with the traditional method.(4) Complete the new model by organic combination of (1), (2) and (3), minimize the energy function, and ultimately get the target contour.At last, the experiment is carried out using Matlab simulation software. The simulation results show that the new model has better performance than PC model, LBF model and LIF model, it can get good segmentation effect in noise environment, and has improved the efficiency and precision, the sensitivity to initial position problem have been greatly eased comparing with LBF model. In our model, the numerical scheme is stable, and can achieve good segmentation results in complex heterogeneous images, it has proved that our model is an effective and superior medical image segmentation method.