The Research And Application Of Image Segmentation Algorithm Based On Fuzzy Clustering

Image segmentation is an extremely important technique of image processing,which plays a key role in pattern recognition and computer vision.With the rapid development of economic society and technology,traditional image segmentation methods are hard to meet the ever-changing requirements in engineering applications.As a popular research area,fuzzy clustering based image segmentation has attracted much more attention in recent years,which has been widely applied to biomedicine,intelligent transportation and target detection.Image segmentation algorithm based on fuzzy clustering evaluates the degree of which a pixel belongs to a certain class using membership,which is different from that based on hard clustering,therefore better representing fuzzy information in an image and avoiding the misclassification of a pixel into only one class.The defects of fuzzy clustering based image segmentation algorithm like noise nonimmunity,local optimum issue and a large amount of calculation,however,result in the limitation to segmentation accuracy and efficiency,thus affecting the quality to image segmentation to some extent.Accordingly,this paper basically pursues the research of image segmentation algorithm based on fuzzy clustering,and then combines it with practical applications.The main work and innovations of this paper are as follows:(1)A robust fuzzy C-means image segmentation algorithm based on adaptive gray-weighted(AGWRFCM)is proposed.Firstly,a local intensity difference measure is defined by this method to reflect the degree of influence of all pixels on local neighborhood.According to the difference of gray level among pixels in a local neighborhood,the weight of each pixel is adaptively adjusted under the control of exponential function,which improves the calculation accuracy of each pixel's intensity.Secondly,an improved distance is adopted instead of the original Euclidean distance to calculate similarity distance between pixels and cluster centers,which guarantees strong robustness to noise and outliers.Finally,the new method based on adaptive gray-weighted as well as the enhanced distance is applied to FCM scheme,thus making image segmentation realized.Experimental results show that AGWRFCM algorithm can get satisfied segmentation result and strong noise robustness.(2)A new FCM method is proposed for brain MR image segmentation and bias field correction(IBFCFLICM).This method firstly fully takes the spatial distance and local intensity information of image pixels into account,then a new local intensity term and a local spatial distance term are presented on which a multiple local information fuzzy factor constructed is based.In view of the redundancy and self-similarity of an image,the non-local image information is subsequently made use of to establish a non-local weight.Next,bias field model is constructed to be incorporated into FLICM algorithm.Lastly,the proposed multiple local information fuzzy factor as well as the non-local weight is embedded into FLICM framework with bias field model for the purpose of brain MR image segmentation and bias field correction.Simulation results of brain MR images prove the effectiveness of IBFCFLICM method.(3)A novel superpixel based FCM method(SPFCM)is proposed for remote sensing image segmentation.Firstly,the SLIC algorithm is used to generate numbers of superpixels in the process of pre-segmentation,which makes individual pixels transformed into superpixels to reduce the running time of subsequent segmentation.Then,the grayscale feature is extracted from each superpixel for further segmentation.At last,a fuzzy clustering scheme based on optimized membership is used for post-segmentation thus getting final segmentation result.Simulation results of remote sensing images show that SPFCM algorithm can make segmentation fast and effectively realized.