No abstract

The transformation of the development of the Arctic is due to modern robotic systems. The use of unmanned vehicles in many industries in the Arctic provides an array of photo and video information. Accuracy of image analysis and pattern recognition is enhanced by image preprocessing. However, the existing binarization algorithms are not universal for images with different distortions and loss of information. The accuracy of binarization algorithms depends on many factors, such as shadows, uneven lighting, low contrast, noise, etc. Images with different characteristics of light and noise are simulated in order to model various lighting conditions on information from digital cameras of robotic systems. The paper investigates global and adaptive image binarization algorithms. The binarized images were obtained using these algorithms and the results of binarized images recognition are compared by an optical character recognition system. An analysis of the comparison results showed that for images made in poor lighting conditions or images with low contrast, or images with high noise levels, adaptive binarization algorithms are better suited. However, in most cases it is not possible to obtain fully correctly recognized images.The paper proposes a new binarization method based on artificial neural networks. The process of creating an artificial neural network is shown, include the parameters for determining the class of a pixel, the adjustment of weights, the architecture of an artificial neural network. A comparison of the proposed artificial neural network with existing image binarization algorithms demonstrate that in most cases the artificial neural network has the result of image processing at the level of adaptive algorithms or higher. The proposed method of images binarization based on the image color characteristics analysis allows to solve image recognition tasks by robotized systems.

No abstract

No abstract

This article discusses the problems of collecting software requirements. The existing computer vision technologies are analyzed and the choice of technology for recognizing handwritten and printed text is justified. The input data for the experiments are described and the results of character recognition for each image category are presented. A method of image preprocessing and recognition of text characters on mobile devices using parallel computing has been developed. On the basis of the proposed method, a prototype mobile application for collecting and digitalizing data obtained during the requirements engineering has been developed.