The work outlines the concept of “post-interpretation” of images and for its algorithmic implementation a model of a post-recognition interpreter is proposed. The recognition results of the initial images entering the recognition system are considered as post-images, and an artificial neural network is used as a post-recognizer. To assess the effectiveness of using the model, it is proposed to use the “expediency criterion” and numerical examples are considered to illustrate the features of its use in systems for recognizing and interpreting images with high risks. Data from preliminary results of experimental testing of a model for recognizing speech commands as part of an interactive operator's manual for performing various tasks and an assessment of its effectiveness are presented.

Keywords: intelligent data processing system, image interpretation, recognition reliability, decision-making criterion, artificial neural network

The paper considers the problem of automatic detection of defects in the geometric parameters of steel ropes of elevator systems using computer vision methods. The features of flaw detection of moving steel ropes based on video sequences are analyzed, associated with the fragmentation of the image of some defects in adjacent frames and the variability of the geometric dimensions of the rope and the characteristics of the defect visible by the camera due to vibrations of the rope during movement. Taking into account the considered features, two algorithms have been proposed: to determine the defect of thickening/thinning of the rope diameter and the defect of undulation. The paper presents the results of experimental testing of algorithms on a special test bench and calculates the reliability indicators of defect detection by the proposed algorithms in the form of precision and recall of detection of each defect individually, as well as the average precision and recall of detection of both considered defects of geometric parameters of the rope as a whole.

Keywords: steel rope defects, instrumental control, non-contact flaw detection, computer vision

The paper presents a solution to the problem of photographic input metrological control of industrial products based on the method of structural approximation synthesis and analysis of one-dimensional structural images. Mathematical models of structural images, algorithms for their synthesis, analysis and interpretation of processing results are described. The results of experimental testing of the proposed algorithms are presented, allowing to evaluate their computational efficiency and reliability of the metrological control procedure.

Keywords: metrological control, image processing, pattern recognition, structural approximation method, dynamic programming