“…The aim of the paper is to describe a formal model for detecting objects in an image by modernized versions of classical cluster analysis methods [ 5 , 6 ] (Otsu’s methods [ 7 , 8 , 9 , 10 ], Ward’s [ 1 , 11 , 12 , 13 ], K-means [ 1 , 14 , 15 , 16 , 17 , 18 , 19 ] and combined splitting/merging clustering methods [ 11 , 20 , 21 , 22 ]), which reduce the image approximation error of approaching an image by its piecewise constant approximations. In this case, the requirement of minimizing the approximation error is to be fulfilled, which, as a rule, conflicts with the heuristic consideration of a priori information about objects.…”