Improvement Of K-means Algorithm And Its Application In Digital Image Segmentation

In the field of image processing,image segmentation is very important for image processing and classification.At the same time,image segmentation which is an important step in digital image processing has played an important role in all aspects of society,such as industrial automation,document image processing,production process control,remote sensing and biomedical image analysis.Therefore,we need to segment these images into different regions and extract the objects of interest.Among different image segmentation techniques,clustering is one of the most important methods and has been widely used in the segmentation of gray-scale images.Many researchers propose and optimize a series of algorithms and methods in clustering-based image segmentation,such as k-means clustering,Fuzzy c-means clustering,mountain clustering and ISODATA algorithm.K-means clustering algorithm is one of the most commonly used algorithm.The K-means algorithm is an unsupervised clustering algorithm that is intuitive,fast and easy to implement.Although this algorithm is very popular,it still has some drawbacks.The main disadvantage is that k-means clustering needs to know the number of clusters in advance,which will reduce its robustness and stability.In addition,the algorithm is more sensitive to outliers and skewness distributions.If the initial cluster centers chosen are outliers,the result of iteration will have a bad influence.n this paper,we improved the k-means according to the above problems.The algorithm also improves the running efficiency of the algorithm and the stability of the algorithm.In this paper,we propose an optimal criterion to overcome the dependence of the initial number of clusters in the k-means algorithm.The optimal criterion can select the best segmentation result with less number of clusters.The optimal criterion overcomes the shortcoming of initialization based on the intra-class and inter-class difference.Many digital images were employed to verify the segmentation results of the optimal criterion.Simultaneously,we have improved the traditional k-means algorithm to find the initial clustering centers efficiently.Experimental results show that the segmented images selected by the optimal criterion have sufficient stability and robustness.In addition,we verify the consistency of results by the objective assessment measures.Simultaneously,we have improved the traditional k-means algorithm to find the initial clustering centers efficiently.We improved the finding of the initial clustering centers by fractile strategy instead of random selection in traditional k-means algorithm.Using fractile points to determine the initial clustering centers increases the efficiency of the segmented images.In order to further verify the performance of the improved k-means algorithm,we apply it to the field of medical image segmentation.Among various medical images,magnetic resonance images(MRIs)play an important role in brain structure recognition and quantitative analysis of medical images.However,due to the influence of imaging equipment and external factors,the accurate segmentation of brain MR images is a challenging task for the relevant researchers.Although there are many unsupervised algorithms acting on the MR images,it must be set up cluster initialization through artificial selection.We use the improved k-means algorithm to determine the best clustering number k and segment the image.Concerning the influence of imaging mechanism,the imaging device,and the individual differences,which may cause the obtained brain MR images having the bias field,we utilize nonparametric nonuniformity normalization(N3)method to revise the bias field.The proposed method can get better segmentation of brain MR images.In addition,we testify our method by comparing it with a variety of evaluation criterions and carry out the quantitative assessment,such as Peak Signal to Noise Ratio(PSNR)and Jaccard Similarity(JS)index.Experimental results show that our method provides higher segmentation accuracy and has higher robustness than other existing methods.