Research Of Image Segmentation Algorithm Based On The Backward Cloud Model

Image segmentation is a key technique in digital image processing and analysis; it is forfurther image analysis, recognition, compression coding and other image processing of theimage pre-processing part, the segmentation directly affects the accuracy of subsequent imageprocessing tasks effectiveness. Image segmentation is a key step from image processing toimage analysis, and occupy an important position in image processing. It is the basis of objectexpression, and has important influence on characteristics measurement; on the other hand,since image segmentation and segmentation based on the target expression, feature extractionand parameter measurements of the original image into a more abstract, more compact form,it can be more high-level image analysis and understanding. Image segmentation points are:the image is segmented a set which have a number of non overlapping, characteristics regions,and extract the interested target; these characteristic region are either meaningful to thecurrent duty, or explain the relationship of them and actual objects or the certain parts ofobject. These characteristics may be the image gradation, the image color, the image texture,and so on.Most of the current image segmentation algorithms are based on deterministic method,however in image information processing there are many uncertainties, moreover thehumanity to the knowledge understanding also frequently reflected uncertainty. Therefore,expression, evaluation and reduce the extraction of target information uncertainty becomes animportant research direction in the process of image segmentation. To deal with uncertaintyproblems, the cloud theory is one kind of new theory which establishes in the traditional fuzzyset theory and in the probability statistics foundation. It uses in the cloud model the clouddrop uncertainty to describe the distance between the elements in spatial data and its coreconcept relationships. The cloud drop’s determination is bigger, data element closer conceptcenter; the cloud drop determination is smaller, the data element more is far away from theconcept the center. It is based on the measurement of the uncertainty, the cloud theory andcloud model method can be better used to image segmentation. Experimental results showthat the cloud theory has a good expression of the uncertainty of concept and reduces theuncertainty of concept hierarchy, so that it can effectively deal with the uncertain problem inimage segmentation. In view of the seed selection of region growing method and uncertainty informationprocessing problems，this paper proposed a method of region growing image segmentationbased on cloud theory. Firstly, it use the global information of image to obtain the seed point,specifically the cloud transform generate the normal cloud model intersection as a seed pointof growing region; and it take the greatly determination law as the growing criterion of theregion growing process; then the reverse cloud algorithm realizes the segmentation regionconversion process from the quantitative pixel set to the qualitative cloud concept; and thecloud synthesis algorithm merge the adjacent regions, finally realizes the uncertainty imagesegmentation based on region growing. The simulation results show that the method can bebetter than other methods of dealing with the uncertainty information of the image, and hasobtained good results.