Technique for Acceleration of Classical Ward's Method for Clustering of Image Pixels

The segmentation task refers to the preliminary stage of image preprocessing. Further object detection, feature recognition, scene analysis and prediction of the situations depends on its results. Modern segmentation algorithms require: the refusal to use of a priori information, the availability of a quality functional for the result assessment, the variable number of segments in the partition of the original image, the linear computational complexity and the adequacy of the segmentation results. Among the methods of the cluster analysis that satisfy most of the requirements listed, the Ward's method is appropriate. But, its high computational complexity prevents its direct application. The purpose of this study is to overcome the excessively high computational complexity of the classical Ward's method. The classical methods of cluster analysis that meet the actual requirements for modern image segmentation algorithms are compared. The choice of the classical Ward's method is justified as well as its advantages and disadvantages are presented. The application of the idea of the reversible computing in image processing is described. The modifications of the computational process of the Ward's method are described. A model flowchart of a sequence of algorithms allowing to bypass the problem of the computational complexity peculiar to Ward's method is proposed. The experimental results of the quality improvement of the conventional segmentation are presented. The proposed model flowchart allows one to bypath the problem of the computational complexity by dividing the process into separate sequential three stages. The model flowchart is suitable for the quality improvement of any traditional segmentation.