Color Image Segmentation Algorithm Based On Quaternions Clustering

Image segmentation is a process to divide an image into non-overlapping, connectedregions, based on the guidelines of similarity and uniformity. It is one of the key technologiesin image processing, which is also the focus and hotspot of discussion and research. For along time researchers have proposed a large number of grayscale image segmentationmethods, but with the rapid development of computer technology and the increase inapplication of color images, color image segmentation methods also have attracted more andmore attention. Color image segmentation can be seen as an extension of the gray-scale imagesegmentation, but the direct application of the gray-scale image segmentation methods oncolor images will not achieve satisfactory effect, which requires researchers to seek more newmethods. This article provides an algorithm of color image segmentation, combined with thequaternion space, based on the histogram, K-means clustering and quaternions. The analysisof experimental results and computer program are given.Firstly, this article analyzes the color image segmentation technology including two parts:color spaces and commonly used color image segmentation methods, in which recently usedcolor spaces and color image segmentation methods are analyzed, and through induction thekey to solving the segmentation problem has been po inted out as to select suitable colorspaces and segmentation methods.Secondly, clustering algorithm has been introduced and detailed analysis and experientalvalidation have been presented for K-means clustering algorithm. To deal with the uncertaintyof the random selection of initial center point of the traditional K-means clustering algorithm,the method of generating the initial focal point through automatic detection of the peak ofhistogram has been proposed. The experimental results show that the center point generatedby the histogram is reasonable, effective, which can reduce the time of clustering iteration, and is better than the randomly selected central point.Thirdly, this article proposes a color image segmentation algorithm based on quaternionsclustering, called Q-means.The analysis of quaternion theory is focused on. Using thequaternion rotation to design a new tonal difference formula, combined with distancemeasurement formula, the difference between the two pixels in the RGB color space can bemeasured, and verificated with the initial focal point through automatic detection. Theexperiment proves that the use of Q-means can get better segmentation results.Finally, the article is concluded and some issues are put forward worthy of furt herin-depth study and discussion.