Research On Image Segmentation Method Based On Image Feature Density Peak Clustering

Recently,there has been paid much more attentions on digital image processing technology.It is gradually involved in all aspects of daily life,which is important for humans to recognize images and make full use of key information derived from images.Image segmentation is an indispensable part of digital image processing,who divides an image into multiple sub-regions and establishes a basis for further investigations.With the development of digital image processing technology,image segmentation technology has made great progress in diverse application fields,such as face recognition,target detection,medical image processing and so forth.Clustering-based segmentation algorithm is one of the more important algorithms in current image segmentation technology.However,the existing widely used clustering methods often have the following problems: First,the accuracy of the segmentation algorithm depends on the prior knowledge used in the clustering process to large extents,and the second is the threshold selection with higher requirements.Thirdly,it needs to manually select the cluster center,and lacks the mechanism of automatically finding and selecting the cluster center.In this study,we proposed an image segmentation method on the basis of image feature density peak search to resolve the above mentioned problems.The work content of this study is as follows:(1)Based on the color features of superpixel extraction images,I analyzed and summarized different image feature extraction methods and then completed the abstract process from irregular color patches to clustering algorithm sample points.(2)Constructed the separation function and proposed an image segmentation method owing to image feature density peak search.In this present study,the local density method was improved and the manual selection of the threshold was removed.The clustering method was applied to each sample point based on superpixels,and constructed the decision graph according to the classification result of each super pixel.Finally,class center was automatically selected relying on the separation function followed by a final segmentation result.(3)Designed experiments to obtain the algorithm of this exploration,and evaluated and summarized the performance of the algorithm.Through the verification of Berkeley segmentation dataset BSDS300,the experimental results of the algorithm model on the common dataset were compared with the results of other commonly used algorithms,indicating that a better segmentation result was achieved when compared with the existing clustering segmentation algorithm.The work innovation of this study:(1)Combining the clustering algorithm in accordance with density peak search,an image segmentation algorithm was proposed.Due to the mentioned method,various methods for image feature extraction were studied.The local density method was changed and the algorithm was elevated by increasing the computing power and fault tolerance.This novel method can help us clearly distinguish the peak density of image features,effectively avoid the defects of manual threshold selection in a large number of existing algorithms,and provide the foundation for the selection of cluster centers.(2)For the first time,the method of adaptively selecting the clustering center was proposed.The partitioning function enabled the algorithm to automatically determine the clustering center in the line with different image features,select different clustering numbers and thus to realize automatic segmentation.