Funct. Mater. 2020; 27 (2): 417-423.

A method for identifying nanoparticles in images obtained with an electron microscope is proposed. Object recognition in the photo is used as algorithms for classic computers, and neural networks, and even systems with the participation of people. The proposed algorithm consists of pre-processing a photograph using a median filter to smooth out brightness drops, separating the image of nanoparticles from the background using a two-stage method. This allows you to reduce the number of gradations of pixel brightness to k, the value of k is selected from the condition of minimizing the function of the "price" C(k). Further, the pixels of the obtained image are classified as pixels of nanoparticles and background pixels from the condition of maximizing the correlation function TC(s). Finally, in the obtained image of the nanoparticles, separated from the background, individual particles are identified. For this, separation and integration of adjacent image regions into separate nanoparticles is performed based on a comparison of the similarity functions δ(i, j) and w(i, j) with threshold values. Further, for the nanoparticles identified in the photo, the areas of their images and perimeters are measured using the proposed algorithm. The results of the proposed method are compared with the results of manual photo processing. The error is about 10 %, which is satisfactory for the goals.