Image Segmentation Based On Clustering

Image segmentation is based on certain criterion and the image is divided into a numberof specific and unique regions. Image Segmentation based on Clustering, which is anunsupervised algorithm, is based on certain criteria through factor analysis.Image segmentation position in various fields is very important, and the ImageSegmentation based on Clustering has played a key role in the field of image segmentation, sothe research about it is very meaningful.In this paper, we will explore the existing image segmentation methods based on clusteringin depth, and suggest improvements based on the methods. Specific contents can be dividedinto the following sections.The first part is the in-depth understanding of existing clustering algorithms and analysisof their strengths, weaknesses and complexity, especially MRSD (Maximize the Ratio of Splitto Diameter) algorithm for a more detailed inquiry. By combining Normalized Cut algorithm,experiments are conducted to verify that the MRSD algorithm can achieve very good results inimage segmentation.The second part presents a MRSD improved algorithm based on K-means preprocessing.Because MRSD algorithm has high time complexity, it’s difficult to apply in the field of high-resolution image segmentation. This paper presents a very broad scope improved algorithmfrom the perspective of reducing the number of pixels. High-resolution image segmentationexperiments show that the improved algorithm can obtain satisfactory results, and improve theefficiency of image segmentation greatly.Finally, we propose a MRSD improved algorithm based on adaptive thresholdpreprocessing. Taking into account the special nature of image, image preprocessing can bebased on a special algorithm. Therefore, based on the characteristics of image, this paperpresents an improved algorithm for image. Compared with MRSD algorithm based on K-means,the improved algorithm can find the regional characteristics of image more accurately insegmentation process. The experiments show that the improved algorithm can obtain verysatisfactory results, too.