Mechanisms of image recovery optimization in the system for recognition and classification of micro-objects

The scientific and methodological foundations for the construction of methods, models, and algorithms of the software complex for visualization, recognition, classification of micro-objects with mechanisms for optimizing image restoration based on the use of structural and statistical redundancy of information have been developed.. Mechanisms for optimizing image recovery based on the principles of using structural-statistical redundancy caused by the repetition of identical frames during stream transmission, interframe difference, and fractal characteristics of images are proposed. The image recovery mechanisms are improved on the basis of multilayer segmentation of the contour, object texture, binary alpha map, creation of metadata arrays, image scaling by bicubic interpolation. Mechanisms for image identification based on a dynamic model of neural networks with component schemes of axon branching, neuronal activity, and revealing their influence on interneuronal connections have been developed. Optimization of NN image restoration is based on the use of process features, axon movement, axon branching, choice of the direction of its growth depending on the concentration of surrounding neurons, emulation of changes in global and additional characteristics of images. A software package has been implemented that includes mechanisms for optimizing image recovery, which has been tested with 538 training sets of medical objects (unicellular microorganisms in the blood).

Перевод пока недоступен