Vol 29, No 1 (2019)

In this paper, we consider algorithms for synthesizing effective distortion-tolerant features and decision rules to recognize weakly formalized objects on large-format bitmap images. In addition, we analyze the effect of binarization of the initial bitmap description on the quality of subsequent automatic recognition.

The properties of doubly stochastic models constructed using a combination of autoregression models with multiple roots of characteristic equations are studied. These models are demonstrated to be adequate to real multidimensional signals; the probabilistic and correlation properties of the simulated signals are studied. Based on the proposed models, a filtering algorithm is developed for doubly stochastic autoregression random fields generated by the models with multiple roots of the characteristic equations. The algorithm is compared to the alternative approaches.

The inspection of power lines is the crucial task for the safe operation of power transmission: its components require regular checking to detect damages and faults that are caused by corrosion or any other environmental agents and mechanical stress. During recent years, the use of Unmanned Autonomous Vehicle (UAV) for environmental and industrial monitoring is constantly growing and the demand for fast and robust algorithms for the analysis of the data acquired by drones during the inspections has increased. In this work, we use UAV to acquire power transmission lines data and apply image processing to highlight expected faults. Our method is based on a fusion algorithm for the infrared and visible power lines images, which is invariant to large scale changes and illumination changes in the real operating environment. Hence, different algorithms from image processing are applied to visible and infrared thermal data, to track the power lines and to detect faults and anomalies. The method significantly identifies edges and hot spots from the set of frames with good accuracy. At the final stage we identify hot spots using thermal images. The paper concludes with the description of the current work, which has been carried out in a research project, namely SCIADRO.

On a comprehensive speech database, speaker recognition characteristics are compared under the usage of various voice-source models. Inverse problems to find a source via vowel speech segments are solved on the base of a special speech-production model and voice-source models (A-source, piecewise-linear source, nonparametric source, and source found by means of the spectral relation method). In the first stage, we find the pulses such that the relative residuals of their segmented and their theoretical analogs computed by means of the speech-production model are less than 0.25. For the selected pulses, a posteriori estimates of the error of their determining are computed and the final selection of the source pulses is performed: for the recognition procedure, we leave only pulses with a posteriori estimates of the error less than the accepted level 0.3. In the space of parameters found for each source model, a statistical model is created for each speaker and the recognition is performed. For the speaker recognition with respect to one vowel, the mean error is approximately equal to 66% for the piecewise-linear source, 61% for the spectral relation method, and 33% for the A-source.

Ground object information on optical remote sensing images is obscured by thin clouds. This paper proposes a ground object information recovery algorithm for thin cloud contaminated optical remote sensing images by combining methods of support vector guidance filtering and transfer learning. Firstly, thin cloud contaminated target images and cloud-free images are decomposed into multidirectional subbands by using multi-directional nonsubsampled dual-tree complex wavelet packet transform (NS-DTCWPT). Then support vector guided filter is applied to remove thin cloud and transfer learning method is used to predict the ground object information on multidirectional subbands. Finally, the processed multidirectional subbands are reconstructed by using inverse multi-directional NS-DTCWPT to obtain the ground object information recovery images. The proposed algorithm combines the advantages of methods of support vector guidance filtering and transfer learning. Experimental results show that the proposed algorithm can effectively remove the thin clouds on the optical remote sensing images and obtain a good recovery effect of ground object information.

This paper presents a method of segmentation of a point cloud into individual objects. The described method is based on using the data on the intensities of the reflected signals obtained by laser scanning. To accelerate the calculations, the point cloud is preliminarily partitioned into non-overlapping sets, superpixels. The objects are segmented by merging the superpixels and taking account of two features: the distance between the histograms of the intensity distribution and the coordinates of the points. To improve the quality of segmentation, three-dimensional filtering using a low-frequency Gabor filter is suggested. The calculation results and an analysis of the effectiveness of the proposed method are given.

Cancer is the major cause of death worldwide. Lung cancer is one of the most common types of cancer and is the main reason of cancer death. Lung cancer is defined as a detrimental lung tumor which is identified by unregulated cell development in the lung tissues. If this disease is not treated at early stages then this unregulated growth can spread into the neighboring tissues and other parts of the body. Detection of lung cancer at its early stage is very difficult because there are very less or may be no symptoms in the early stages of this disease and most of the cancer cases are usually diagnosed in its subsequent stages. Treatment of lung cancer in the early stages can improve the survival rate. And for this purpose it is an essential task to detect the lung cancer at its early stages. In this paper we have presented a comparative analysis of various image segmentation techniques for the detection of lung cancer. These methods include Thresholding methods, Marker Controlled watershed Segmentation, Edge detection and PDE based segmentation techniques.

Watermarking on multimedia gives more attention in research society. In this non-blind watermarking approach a watermark is reshaped and grouped into odd and even images. Next, wavelet is imposed on an original image and to enrich the robustness of the technique, watermarking process is enabled only in the upper diagonal regions of middle level wavelet coefficients. Further, the backward process is taken place to obtain the watermarked image. To prove ownership, original and watermarked image have undergone the same operation and acquire the copyright information. Experimental results indicate that the quality of the watermarked image is better also this algorithm is robust against filtering and geometrical attacks.

An approach to subpixel corner and edge detection in images is described. The approach is based on algebraic moments of the brightness function describing a halftone image. For an ideal two-dimensional L-corner edge, we consider a model with the following four parameters: the coordinates of the corner vertex, the orientation and the degree measure of the corner, and the brightness values from both sides of the edge. A particular case of the angular model is a linear model that describes a linear edge. To obtain all parameters of the model, six algebraic moments are used. To compute the moments rapidly, masks are used. Based on the angular model, we propose an algorithm that makes it possible to quickly refine the coordinates of the corner and edge points in the image with subpixel accuracy. The use of integral characteristics increases the resistance to various kinds of noises. We perform experiments to confirm the efficiency of the proposed approach.

The methodology of forecasting using neural networks for modified data with original values replaced by discrete ones was successfully applied in previous authors’ works for mixture moments approximated with method of moving separation of mixtures. Previously the uniform architecture has been used for all analyzed time-series. This paper proposes an improved type of neural network architecture with grid-search hyper-parameter tuning for expectation, variance, skewness and kurtosis. It allows increasing the value of prediction accuracy for some combinations of forecast periods and number of gaps up to 99.7%. Numerous tables and plots are presented to better demonstrate the results.

The author’s name R. A. Paringev should read R. A. Paringer

The author’s name A. V. Soifer should read V. A. Soifer