A Boundary Stability Constrained Multi-object Segmentation Method By Multi-region Thresholding Based On Intensity And Space Class Uncertainty

With the improvement of people’s living standards,the public’s demand for a healthy life is also growing.Nowadays,medical image segmentation has become one of the important research directions in the field of health care,and its segmentation effect has an important impact on medical diagnosis,scientific research and other matters.However,medical image segmentation is different from traditional image segmentation.Affected by the mechanism and biological characteristics,medical images often suffer from problems such as uneven intensity distribution,heavy noises,and fuzzy boundaries between biological tissues,which makes it difficult for classical thresholding methods to receive accurate results in complex medical image samples.The particularity of the application scene of medical image segmentation determines the key and importance of segmentation accuracy.Traditional methods are not ideal in terms of accuracy,and manual segmentation methods are far from meeting the requirements in terms of efficiency,so how to segment medical images efficiently and accurately is a great challenge.In recent years,with the development of artificial intelligence technology,medical image segmentation methods based on deep learning have emerged one after another,but such methods require a large number of sample training and time-consuming.In contrast,the threshold-based segmentation method is simple,efficient and does not require training samples,so it is widely favored by researchers.This paper focuses on the threshold segmentation method of image gray information and spatial information fusion under complex sample conditions and the local threshold segmentation framework.The main work of this paper is as follows:(1)Based on the expansion of the traditional class uncertainty theory,the unstability of the boundaries theory and the context-based entropy theory are proposed.On this basis,the concept of spatial class uncertainty and stratified gradient are constructed,and finally combined into the region score function to achieve segmentation.In other words,a new threshold segmentation measure is proposed by fusing image intensity information and spatial information.This method is abbreviated as MSSS(the Maximization of Spatial Class Uncertainty-and Stratified Gradient-based Score method).Experiments show that compared with the traditional methods,the proposed method has better robustness to noise,uneven intensity distribution,fuzzy boundaries and other interference,and the segmentation results are also better.(2)Aiming at solving the problem that the global thresholding methods cannot deal with complex images,a local thresholding segmentation framework is proposed.In this framework,the image is divided into different regions by selecting the base point in the image and diffusing sub-regions by distance transform.Then,the selection of sub-regions is further optimized by alternating optimization algorithm,and finally,the optimal threshold value is selected in each sub-region.Due to traditional local thresholding methods often define the region selection logic artificially,there may be some problems in the segmentation results such as noise or poor continuity.The proposed method uses the base point and distance transformation to make the sub-regions diffuse freely,avoiding this problems.After experiments on multiple public data sets and real data,it can be concluded that,compared with other excellent methods,the proposed method has good performance in terms of accuracy,time complexity and repetition stability.