The ascending approach to segmentation of scanned documents in the area of background, text, and photographs is considered. In the first stage, the image is divided into blocks. For each block, a series of texture features is calculated. On the basis of these features, the type of the block is determined. Various positions and sizes of blocks, 26 texture features, and 4 algorithms of classification of blocks were considered. In the second stage, the type of block was corrected on the basis of the analysis of neighboring regions. For estimating the results, the error matrix and the ICDAR 2007 criterion are used.
Convolutional Neural Networks (CNNs) and Deep Learning (DL) revolutionized numerous research fields including robotics, natural language processing, self-driving cars, healthcare, and others. However, DL is still relatively under-researched in physics and engineering. Recent works on DL-assisted analysis showed enormous potential of CNN applications in electrical engineering. This paper explores the possibility of developing an end-to-end DL analysis method to match or even surpass conventional analysis techniques such as finite element analysis (FEA) based on the ability of CNNs to predict the performance characteristics of electric machines. The required depth in CNN architecture is studied by comparing a simplistic CNN with three ResNet architectures. Studied CNNs show over 90% accuracy for an analysis conducted under a minute, whereas a FEA of comparable accuracy required 200 h. It is also shown that training CNNs to predict multidimensional outputs can improve CNN performance. Multidimensional output prediction with data-driven methods is further discussed in context of multiphysics analysis showing potential for developing analysis methods that might surpass FEA capabilities.