R.A. Tomakova, S.A. Filist, A.A. Naser

The article considers indistinct neural-net technology for automated analysis of microscopic photos of complex-structured images-image of blood smear photo has been taken as an example. There has been worked out special software realized in medium Matlab 7.10, which is presented as interactive co-operation of the software with a person, making a decision, and is fulfilled by means of interface windows and floating-up menus. Functional feature of which is in sequential inclusion of each module into technological process of analysis. Module of software segmentation consists of two program blocks: 1) block of neural-net classification for image pixels; 2) block of image segments morphological analysis. The software of the block for neural-net classification of image pixels in the educational process carries out formation of neural networks models for image segmentation. Basic technological operations here are: choice of mask (window) sis; choice of color channels (R, G or B); formation of educational sample; adjustment of neural network parameters; teaching of neural network; saving of neural network model in data base. During the process of image segmentation program module carries out sampling of neural networks models base more suitable to the image under investigation according to chromatic. This is fulfilled both in interactive and automatic modes. Preparation of segmented binary image by means of module of morphological processing is fulfilled after the procedure of the chosen image segmentation. Results of morphological operators work are given in figures. Binary image, necessary for work of program module for formed blood elements classification, comes out as a result of carrying out the sequence of morphological operators. Given software allows to realize the process of segmentation and classification of complex-structured images on the basis of complete technological cycle for synthesis of multilevel models of neural networks. It also includes processes for educational samples formation, computation of neural networks parameters and determination of diagnostic efficacy of obtained decisive modules.