Research On Image Segmentation Based On Fuzzy Clustering Algorithm By Multi-pixel Fitting

Image segmentation is a crucial process in machine vision.Image segmentation algorithms can be divided into two types: supervised and unsupervised.Supervised segmentation algorithms need to use labeled samples for training and learning,but unsupervised segmentation algorithms directly process the data.Fuzzy clustering algorithm(FCM)is a classic unsupervised segmentation algorithm,but it is susceptible to image noise and outliers.Therefore,the existing improved FCM algorithms introduces spatial information to enhance the ability of the FCM to resist image noise and outliers.Although part of the improved algorithms can achieve the ideal segmentation effect,their computational complexity is high,which may lead to lack of practicability.Aiming at the problem that the existing improved FCM algorithm cannot have both strong robustness and high efficiency at the same time,the research of image segmentation algorithm based on multi-pixel fitting fuzzy clustering is carried out in paper.The main research work of the paper is as follows:(1)Aiming at the shortcomings of existing FCM improved algorithms that cannot have strong robustness and high efficiency,an efficient fuzzy clustering algorithm based on multi-pixel fitting(FCM-MF)is proposed.In FCM-MF,a filter window containing multiple pixels,its corresponding generalized neighborhood window,and a efficient way to obtain spatial information through multi-pixel fitting are innovatively defined.Using the FCM-MF and existing improved algorithms to conduct comparative experiments on the synthetic and real images.Experimental results show that the FCMMF are better than the comparison algorithm on the synthetic image,verifying the proposed FCM-MF has extremely robust.At the same time,the segmentation time is less than the seven comparison algorithms,which verifies that the proposed FCM-MF algorithm has higher segmentation efficiency.It also has efficient and fast segmentation performance in the comparison experiment of real images,but it is also found that the FCM-MF needs to be strengthened in the edge details of some images.(2)In order to enhance the segmentation performance of the FCM-MF at the edge of images,a variable window strategy is introduced into the FCM-MF,and the FCMVMF is proposed.The proposed strategy includes the following three steps: First,obtain the label of input image using FCM;Then,the ratios of the number of pixels belong to different class to the number of all pixels in the generalized neighborhood window are obtained;Finally,the maximum value of the obtained ratios is employed as an indicator to reduce the size of generalized neighborhood window adaptively.The proposed FCMVMF algorithm will be compared and tested on the synthetic image and 170 real images.The experimental results show that the average DICE of the segmentation results of the FCM-VMF on the noise-added and noise-free images is close to the optimal FLICM algorithm,and the segmentation time is the least,which proves the FCM-VMF is robust to noise and can ensure the efficiency of segmentation.(3)The proposed FCM-MF and FCM-VMF algorithms are compared on real images.And,the FCM-MF algorithm and FCM-VMF are analyzed in segmentation efficiency and image detail information.Experimental results show that both algorithms can effectively remove image noise,ensure segmentation efficiency and obtain good segmentation results.Among them,the segmentation efficiency of the FCM-MF will be slightly higher than FCM-VMF.But in the edge area of some images,the edge information of the FCM-VMF will be more complete,accurate and credible.(4)The FCM-VMF is extended to COVID19 CT images.Firstly,the FCM-VMF algorithm is used to segment the COVID19 CT images.Then,the lungs and suspected lesion areas are extracted according to the segmentation results.Finally,use labveiw software to systematically design the extraction process of lungs and suspected lesion areas,and obtain a COVID19 CT imaging lung lesion diagnosis system.In the above process,the FCM-VMF algorithm can quickly and effectively extract the lungs and suspected lesion areas in the CT images of new coronary pneumonia,highlighting the good applicability of the FCM-VMF algorithm.