Computer Optics

522-530

Authors: V.V. Kotlyar, E.G. Abramochkin, A.A. Kovalev, A.G. Nalimov

It is shown theoretically that an astigmatic transformation of an edge dislocation (straight line of zero intensity) of the ν-th order (ν=n+α is a real positive number, n is integer, 0<α<1 is the fractional part of the number) forms at twice the focal length from a cylindrical lens n optical elliptical vortices (screw dislocations) with a topological charge of –1, located on a straight line perpendicular to the edge dislocation. Coordinates of these points are zeros of the Tricomi function. At some distance from these vortices and on the same straight line, another additional vortex with a topological charge of –1 is also generated, which moves to the periphery if α decreases to zero, or approaches n vortices if α tends to 1. In addition, at the periphery in the beam cross-section, a countable number of optical vortices (intensity zeros) are formed, all with a topological charge of –1, which are located on diverging curved lines (such as hyperbolas) equidistant from a straight line on which the main n intensity zeros are located. These "accompanying" vortices approach the center of the beam, following the additional "passenger" vortex, if 0<α<0.5, or move to the periphery, leaving the "passenger" next to the main vortices, if 0.5<α<1. At α=0 and α=1, the "accompanying" vortices are situated at infinity. The topological charge of the entire beam at fractional ν is infinite. The numerical simulation confirms theoretical predictions.

537-546

Authors: D.A. Savelyev

The focusing features of a super-Gaussian beam, as well as Laguerre-super-Gaussian (1,0) modes with radial and circular polarizations on ring gratings (direct and inverse) with a variable height of individual relief rings were investigated in this paper. The change in the height of the relief from the maximum height in the center to the minimum at the edges of the element (direct ring grating) is considered and the reverse case, when the maximum height of the relief was at the edges and the minimum in the center (reverse ring grating). The comparison was carried out with the action of a diffractive axicon with a comparable grating period. Numerical simulation was carried out by the finite differences in the time domain method. It was shown that the direction of change in the height of the relief rings of elements significantly affects the diffraction pattern in the near zone. In particular, for a super-Gaussian beam, the use of a direct ring grating made it possible to reduce the size of the focal spot by 18.7% for radial polarization ('sigma'=5 μm), the use of a reverse ring grating led to a decrease in the focal spot by 36.9% ('sigma'=12.25 µm) in comparison with the action of a diffractive axicon. It was also shown in the paper that when the reverse ring grating is illuminated by the Laguerre-superGauss (1,0) mode with circular polarization, the formation of a narrow and extended region of reduced intensity on the optical axis (optical trap) is observed.

555-560

Authors: V.S. Petrakova, A.S. Tsipotan, V.V. Slabko

One approach for the formation of structures with complex geometries at the nanoscale is the step-by-step assembly. In this case, it is necessary to be able to estimate the time required to establish orientational equilibrium for a preformed pair of particles. This process is statistical in nature and depends on the mechanism of interaction of the ensemble with the external field. The orientation of particles in an alternating field is associated with certain relaxation times, which depend on the viscosity and temperature of the medium, as well as on the geometric structure of the samples. This paper proposes an mathematical model of the process of establishing the distribution of nanoparticles pairs orientations taking into account the friction force, thermal motion, and the orienting laser field. A statistical orientation distribution was obtained for CdTe particles in the field of moderate laser radiation, and the average time for establishing orientational equilibrium was estimated.

567-589

Authors: V.V. Arlazarov, E.I. Andreeva, K.B. Bulatov, D.P. Nikolaev, O.O. Petrova, B.I. Savelev, O.A. Slavin

This paper analyzes the problems of document image recognition and the existing solutions. Document recognition algorithms have been studied for quite a long time, but despite this, currently, the topic is relevant and research continues, as evidenced by a large number of associated publications and reviews. However, most of these works and reviews are devoted to individual recognition tasks. In this review, the entire set of methods, approaches, and algorithms necessary for document recognition is considered. A preliminary systematization allowed us to distinguish groups of methods for extracting information from documents of different types: single-page and multi-page, with text and handwritten contents, with a fixed template and flexible structure, and digitalized via different ways: scanning, photographing, video recording. Here, we consider methods of document recognition and analysis applied to a wide range of tasks: identification and verification of identity, due diligence, machine learning algorithms, questionnaires, and audits. The groups of methods necessary for the recognition of a single page image are examined: the classical computer vision algorithms, i.e., keypoints, local feature descriptors, Fast Hough Transforms, image binarization, and modern neural network models for document boundary detection, document classification, document structure analysis, i.e., text blocks and tables localization, extraction and recognition of the details, post-processing of recognition results. The review provides a description of publicly available experimental data packages for training and testing recognition algorithms. Methods for optimizing the performance of document image analysis and recognition methods are described.

596-602

Authors: R.A. Paringer, A.V. Mukhin, N.Y. Ilyasova, N.S. Demin

Advances in the neural networks have brought revolution in many areas, especially those related to image processing and analysis. The most complex is a task of analyzing biomedical data due to a limited number of samples, imbalanced classes, and low-quality labelling. In this paper, we look into the possibility of using neural networks when solving a task of semantic segmentation of fundus. The applicability of the neural networks is evaluated through a comparison of image segmentation results with those obtained using textural features. The neural networks are found to be more accurate than the textural features both in terms of precision (~25%) and recall (~50%). Neural networks can be applied in biomedical image segmentation in combination with data balancing algorithms and data augmentation techniques.

612-620

Authors: A.Y. Denisova, V.V. Fedoseev

The article proposes a feature system designed to detect presentation attacks on facial authentication systems. In this type of attack, an attacker disguises as an authorized user using his image. The feature system assumes the possibility of using one or more special imaging sensors in addition to the basic RGB camera (thermal cameras, depth cameras, infrared cameras). The method has demonstrated a low error rate on the WMCA dataset, while experiments have shown its ability to remain effective in the case of the lack of training data. The comparative experiments carried out showed that the proposed method surpassed the RDWT-Haralick-SVM algorithm, and also approached the results of the MC-CNN algorithm, based on deep learning, which requires a significantly larger amount of training data.

628-633

Authors: V.Y. Kovalev, A.G. Shishkin

In this paper, a multiclass image semantic segmentation problem was solved. For analysis, images of the intracytoplasmic sperm injection process were used. For training the neural network, 656 frames were manually labelled. As a result, each pixel of the images was assigned to one of four classes: microinjector, suction micropipette, oolemma, background. An analysis of modern approaches was carried out and the best architecture, encoders, and hyperparameters of the neural network were selected experimentally: the convolutional neural network FPN (feature pyramid network) with the resnext101 encoder having a depth of 101 layers with 32 parallel separable convolutions. The developed neural network model has allowed obtaining the segmentation efficiency of IOU=0.96 at the algorithm speed of 15 frames per second.

643-649

Authors: M.V. Gashnikov, A.V. Kuznetsov

We investigate algorithms for detecting artificial fragments of remote sensing images generated by adversarial neural networks. We consider a detector of artificial images based on the detection of a spectral artifact of generative-adversarial neural networks that is caused by a layer for enhancing the resolution. We use the detecting algorithm to detect artificial fragments embedded in natural remote sensing images using an adversarial neural network that includes a contour generator. We use remote sensing images of various types and resolutions, whereas the substituted areas, some being not simply connected, have different sizes and shapes. We experimentally prove that the investigated spectral neural network detector has high efficiency in detecting artificial fragments of remote sensing images.

659-665

Authors: A.V. Kamenskiy

The development of modern technologies in the field of image formation leads to an increase in the size of the generated images, as a result the question of reducing the processing computational costs arises, and this is an important factor in the creation of real-time systems. The study provides a description of high-speed recursive-separable filters for improving the quality of images, which, due to the peculiarities of their implementation, can reduce the number of computational operations required for the image processing process. This type of filters is obtained from two-dimensional linear digital filters, which are modified by applying recursive and separable properties to them. The MATLAB environment computing method for implementation of these filters is described. An extensive performance research of the developed filters has been carried out at various sizes of the test image and on various experimental installations. The comparison with the classical two-dimensional convolution method of the developed filters is demonstrated, and it shows the time gain required for the image processing. The results obtained can be applied in biomedical image processing systems or in vision systems working in heavy weather conditions.