MULTISPECTRAL IMAGE COMBINING AS A METHOD OF INFORMATION CONTENT INCREASING AT BINARY SEGMENTATION

Abstract

Background. One of the biggest tech trends in manufacture and equipment engineering today is the desire to automate processes in order to increase their efficiency and minimize the influence of the human factor. A promising direction for the implementation of this concept is computer vision methods based on the automation of image processing and analysis. Machine learning allows systems, such as artificial neural networks, to detect, recognize and classify objects. All this has become possible due to the development of image processing methods, which include binary segmentation. The information content increase of the result with this data processing method is possible using the multispectral complexing algorithm, which will improve the efficiency of machine vision systems and automated image processing.Objective. The purpose of the paper is the investigation of applicability of combining multispectral images methods to increase the information content and quality of the result of data processing based on binary segmentation.Methods. For investigating of the combining images methods the software environment implemented on the C# programming language basis was used with mathematical apparatus of binary segmentation. The results analysis was made using methods of statistical data processing.Results. Effectiveness estimates of the following multispectral image combining methods applying to improve the quality of binary segmentation: averaging, maximum, interlaced combining and interlaced combining of maxima are obtained. The research was made both for images obtained at favorable conditions and in the presence of various types of noises, which decrease the information content of images in different spectral ranges.Conclusions. Using multispectral image combining methods allows increasing the information content of the binary segmentation result up to 15%. Particularly significant quality improvement is manifested in presence of noises on various spectral range images, such as flare or low light.