| Image segmentation is a basic operation in digital image processing, which, as an im-portant theoretical problem, has been investigated from different aspects. Actually,there are two kinds of digital images: gray (or black and white) images and color im-age, and the pixels of the two kinds of digital images are represented by 1-D and 3-Ddata, respectively. Since their dimensions are different, we generally use the differentalgorithms to solve the image segmentation problems for the two cases. With the devel-opment of the automatic image processing, more and more interest has been focused onthe unsupervised or automatic image segmentation algorithm which segments an imageautomatically without knowing any prior knowledge of the image. That is, the unsu-pervised image segmentation algorithm should identify all the objects as well as thebackground in the images and segment them automatically. In this way, if a clusteringalgorithm is applied on the image data, it should determine the number for the clustersand locate the centers of these clusters. In fact, the DSPRCL algorithm is just a typicalclustering algorithm of this kind, which has been established recently. In this paper,the DSRPCL based unsupervised image segmentation is deeply studied. On the whole,the DSRPCL algorithm can automatically determine the correct number of the clustersin the image data, locate the centers of the clusters, and make the image segmentationefficiently, which meets the requirements of unsupervised image segmentation. Seg-mentation results on several gray and color images shows that the DSRPCL algorithmis applicable and efficient for unsupervised image segmentation algorithm. And be-cause of the singular point problem, HSI color space is not suitable for DSRPCL basedsegmentation. Moreover, since the local 3D histogram method is difficult to get thereasonable bins on a color image, we propose a DSRPCL based automatic bin-dividingmethod in this paper, which improves the image segmentation considerably. |
PDF Full Text Request