Методика ускорения классического метода Уорда для кластеризации пикселей изображения

Segmentation refers to the stage of pre-processing images. Further selection of objects, recognition of features, analysis of scenes and prediction of situations depend on its results. The requirements to modern segmentation algorithms are as following: refusal to use a priori information, availability of a quality functional for evaluation of the result, a variable number of segments in decomposition of the original image, linear computational complexity and adequacy of the segmentation results. The Ward method is appropriate among the methods of cluster analysis that satisfy most of the above requirements. However, its high computational complexity prevents its direct application. The aim of the study is to find a way to overcome the excessively high computational complexity of the classical Ward method. The methods: classical methods of cluster analysis that meet the current requirements for modem image segmentation algorithms are compared. The choice of the classical Ward method is substantiated. Its advantages and disadvantages are given. The application of the idea of reversible operations in image processing is described. Results: a number of some modifications of the computational process of the Ward method are given. A typical block scheme of a sequence of algorithms is proposed, that allows to circumvent the problem of computational complexity characteristic of the Ward method. Experimental results are presented to improve the quality of traditional segmentation. Practical significance: the proposed scheme allows to circumvent the problem of computational complexity by dividing the processing course into three consecutive stages. The scheme is suitable for improving the quality of any traditional segmentation.