On Clustering And Fuzzyness Based Image Segmentation And Applications

Image engineering summarizes the overall image-related technologies. It is broadly divided intoimage processing, image analysis and image understanding. Image processing is one of the low-leveloperations, including image segmentation, object separation, feature extraction and parametermeasurement. The related processes in image processing provide the basic conditions for theimplementation of image analysis and image understanding.In the field of image processing, on the one hand, cluster analysis can be seen as thepreprocessing step of image processing algorithms. Image data can be preprocessed throughclustering algorithms so that the image processing efficiency may well be improved; On the otherhand, image segmentation is the process of partitioning an image into different regions with somespecific properties and extracting the interested objects. Segmentation accuracy determines thesuccess or failure of image processing and analysis. Therefore, cluster analysis and imagesegmentation are of great importance for object recognition, image retrieval, and computer vision.As for the importance of cluster analysis for image processing, firstly, we have a good researchfor clustering algorithms and propose non-Euclidean metrics based clustering method. Mosttraditional clustering algorithms adopt Euclidean distance, so that they are just appropriate for convexor sphere data set. However, non-Euclidean metrics, such as kernel metric, Mahalanobis distance andthe metric based on the shortest weighted path, based clustering algorithms have the benefit of notconfining the space shape of data set. Experimental results demonstrate that the application scope ofthese clustering algorithms has been extended by adopting non-Euclidean metrics. Furthermore,image segmentation and classification methods based on clustering are introduced in this paper.The traditional fuzzy connectedness method has some problems, such as segmentationperformance is largely determined by the specified fuzzy relation, sensitive to noise, difficult todetermine an appropriate threshold in the case of multiple seeds, etc.. While these problems can beovercome by a new proposed image segmentation method, in which the density properties of pixelsbased on a neighborhood density index are applied. In this paper, we propose a novel way to capturethe global fuzzy connectedness, and reasonably define the dense and non-dense set of an image.Taking the dense set and non-dense set into account, this paper presents related algorithms for imagesegmentation. Extensive evaluations and analysis demonstrate the utility of such novel approach.