Research On FCM Image Segmentation Algorithm Based On Spatial Constraints

As the first step of image processing,image segmentation is the basis of image analysis and pattern recognition.Medical image segmentation technology for medical imaging plays an important role in the development.However,due to the complexity of the organizational structure of medical images and the high accuracy of segmentation results,as well as the interference of medical imaging equipment and technology,the segmentation of medical images is still a challenge and difficulty,thus many segmentation technologies are derived.Image segmentation techniques mainly include image segmentation based on threshold,image segmentation based on edge,image segmentation based on region and image segmentation based on clustering.In clustering algorithms,the fuzzy c-means(FCM)algorithm is the most commonly used in medical image segmentation.It can obtain effective clustering results by using special fuzzy attributes,and the implementation is simple.But the traditional FCM algorithm does not perform well on magnetic resonance imaging(MRI)data with noise and intensity inhomogeneity(IIH).It is easily interfered by noise and other factors such as IIH,which will eventually lead to inaccurate final clustering results and unnecessary errors in subsequent image analysis and pattern recognition.This will affect the final diagnosis.Therefore,it is particularly important to improve the traditional FCM algorithm to avoid the influence of these disturbances.This paper mainly studies medical image segmentation technology from the perspective of local spatial information of images,so that accurate results can be obtained even under the interference of these factors.The specific contents are as follows:(1)In view of the disadvantages of FCM algorithm in brain image segmentation due to noise and IIH interference,resulting in inaccurate segmentation,this paper proposes a Gamma-enhanced FCM algorithm(GcsFCM)based on spatial information to solve this problem.Firstly,the algorithm introduces the Gamma enhancement algorithm as preprocessing to enhance the details of the image.Secondly,the neighborhood validity function is introduced into the local spatial information,which describes the degree of participation in generating the local membership and building the cluster.This can improve the robustness of noise and IIH.Finally,this function is combined with spatial information to form a weighted membership function.Experimental results on four image volumes with noise and IIH show that the proposed GcsFCM algorithm is more efficient and robust to noise and IIH than the FCM,sFCM and csFCM algorithms.(2)In order to further improve the segmentation robustness of FCM algorithm to noise and IIH brain MRI images,this paper proposes a FCM algorithm(TcsFCM)that combines spatial information and transfer learning.Firstly,transfer learning is applied to gray scale medical brain images in this paper,and the information in the source domain is introduced as prior knowledge into the segmented images(target domain).Secondly,spatial information is introduced to reduce noise and IIH effects.Meanwhile,in order to distinguish the influence of neighboring pixels on the center pixel,the algorithm in this paper assigns different weights to neighboring pixels based on their distance from the center pixel.Finally,the transfer learning function is introduced as a penalty to modify the iteration segmentation results.Experimental results on image volumes with different noises and IIH show that compared with some improved algorithms,our algorithm can effectively avoid the influence of noise and IIH on image segmentation accuracy.