We describe a method for digital image analysis, which is based on the representation of an image by the oriented graph. Vertices correspond to image pixels, edges connect nearest neighbors. We assign a measure to all edges so that to obtain Markov chain on the graph. In accordance with the initial measure distribution the stationary flow is constructed and weighted entropy is calculated. The algorithm is implemented both for the base case (vertex corresponds to one pixel) and the optimized one –- vertex corresponds to a cell of the image partition. The choice of the maximum allowed cell size depends on the image structure and may be obtained experimentally – comparing the weighted entropy values and run times for base and optimized variants. The results of calculations for some classes of biomedical preparations images are given. The described optimization reduces run time in 3-4 times.