Section 2 - Track 2

On this page, talks in the format of posters are posted. They can be viewed during the whole time of the Conference. To send your question to the authors, please follow to Miro and write your question in the comments. You can also use the feedback form on our website. Authors will answer questions using Miro comments or via email. To get manual go to Instructions (eng / rus).

Anna Egorova
The effectiveness of image filtering by superpixel representation
The paper presents a superpixel-based image filtering algorithm for additive white Gaussian noise (AWGN) reduction. The algorithm processes an image by connected homogeneous regions of small size (superpixels). Each superpixel is restored using the least squares method. The mean square error (MSE) between a reconstructed image and an ideal image provided by the proposed algorithm is compared with the MSE provided by the Wiener filter. The experimental part shows that the proposed superpixel filtering algorithm outperforms the Wiener filter, providing lower MSE values.

Evgeny Myasnikov
Assessment of camera orientation in Manhattan scenes using information from optical and inertial sensors
In this paper, the assessment of the camera orientation is performed under two main limitations. The first one is the restriction of the class of analyzed scenes to Manhattan scenes only. The second limitation is the presence of an accelerometer on a mobile device. To assess the characteristics of the proposed solution, a dataset was prepared containing both the photographs and data from the accelerometer, as well as information about the true orientation of the device. Experimental studies were conducted using the prepared dataset.

Yulia Ganeeva and Evgeniy Myasnikov
Iris segmentation in an image using a convolutional neural network architecture U-Net
The accuracy of segmentation plays an important role in the methods for person identification by iris images. In this paper, we study the iris segmentation method based on the convolutional neural network of the U-Net architecture. As part of the research, manual segmentation of the images of the used data set was performed, the optimal network training parameters were determined and the segmentation quality was evaluated. All studies in the paper were conducted using the open MMU Iris Image Database. The results showed that the studied approach provides high precision segmentation of the iris images.

Yuliya Vybornova and Aleksey Maksimov
A comparative study of restoration techniques for images defined by chaotically scattered point set
Fast interpolation algorithms suggest that the value at a particular element of the image is calculated based on some neighbourhood, the choice of which can significantly affect the interpolation result. In this paper, we consider some well-known interpolation methods and research on how the choice of reference image elements affects their quality. Results of the experimental research, such as the dependencies of RMSE on the noise proportion in test images, are presented, as well as computational complexity assessment.

Aleksey Maksimov and Mikhail Gashnikov
Differential method of multidimensional signals compression based on the adapted parameterized interpolation algorithm
In this paper, parameterized algorithms of multidimensional signal interpolation are adapted for use as part of differential compression methods. These methods are based on the efficient coding of quantized differences between the initial and interpolated signal samples during sequential signal scanning. The proposed interpolators are based on the classification of signal samples and the use of various interpolation formulas within the classes. The sample classifier and its training procedure and a set of interpolating functions for the compression method are described. The results of experimental research on real multidimensional signals confirm that the use of an adapted parameterized interpolator leads to an increase in the efficiency of the differential compression method

Yuliya Podgornova and Sultan Sadykov
Increasing the contrast of mammograms containing breast cancer regions on the background of fat involution using wavelet transformations
According to the International Agency for Research on Cancer, in 2018, Russia ranks fifth in cancer mortality. Every sixth woman was diagnosed with breast cancer. The only way to diagnose this disease is the timely passage of professional examinations. For women older than 40 years, a mammary gland examination with mammography is provided. Most of the state medical institutions are equipped with outdated mammographic systems, which often do not allow obtaining images of the required quality. The presence of any type of mastopathy also complicates the diagnosis, as a result of which, on the screening mammogram, it is easy to skip the nucleation of an oncological tumor. The paper proposes the use of wavelet transforms to increase the contrast of mammographic images both against the background of fat involution, and against the background of fibrocystic disease and adenosis. The aim of the work is to study the approach to increasing the contrast of mammograms containing breast cancer on the background of fat involution, based on

Nikolay Glumov and Mikhail Gashnikov
Algorithm for optimizing quantization scales by an arbitrary quality measure
The paper considers the problem of constructing quantization scales that are optimal according to a specified criterion and satisfy a specified constraint. The mathematical statement of the optimization problem is considered. An algorithm is proposed for constructing quasi-optimal quantization scales that approximate the optimal scales with specified accuracy, subject to the constraint. Requirements to the optimization criterion and the constraint are formulated, which ensure the operability of the algorithm. Computational experiments are being carried out to construct quasi-optimal scales. The experimental results confirm the advantage of the constructed scales over the known ones.

Anna Egorova, Victor Fedoseev
Semi-fragile watermarking algorithm for H.264 video protection
The paper presents a semi-fragile watermarking system designed for H.264 video authentication. The proposed system embeds a semi-fragile watermark into the integer DCT coefficients of each keyframe of a video at the H.264 quantization step. The system localizes the modified video regions of size 4×4 with high accuracy. The experimental part demonstrates the influence of the embedding process on the visual quality of the protected video, as well as the watermark extraction error for different H.264 quality parameter values.

Olga Omelchenko and Victor Fedoseev
A method for protecting images from changes with informative fragment recovery option
The paper proposes a method of protecting images from unauthorized localized changes with the possibility of an approximate recovery of the most significant fragments that are automatically selected at the preliminary stage. Protection is provided by embedding semi-fragile digital watermarks, which allow us to construct a mask of changes. The watermark also contains the information necessary to restore a distorted image. Due to the proper choice of the watermarking system, as well as the data encoding order, the proposed method is able to save a protective watermark when cropping the image, as well as when compressing it with the JPEG algorithm. As an example, the application of the developed method for protecting images of the road situation from unauthorized adjustment of car license plate zones is considered. The results of testing the method has shown its efficiency for solving the stated problems.

Evgeny Myasnikov
Hyperspectral data dimensionality reduction using nonlinear autoencoders
The known feature of hyperspectral images is a high spectral resolution, which allows to identify materials and classify objects in images with high accuracy. However hyperspectral images contain substantial redundancy, which can be eliminated with the aid of dimensionality reduction techniques. In this paper, we propose and study several dimensionality reduction techniques based on the pretraining the encoder-decoder neural network with the results of the nonlinear mapping and principal component analysis techniques. The experiments performed on an open dataset show that the proposed techniques both provide the discriminative low-dimensional features and allow to reconstruct source hyperspectral data with little error.

Mikhail Gashnikov
Adaptive interpolation for heterogeneous multidimensional signals fusion
An adaptive parameterized interpolator is proposed to solve the problem of heterogeneous multidimensional signals fusion, based on automatic switching between several interpolating functions at each point of the signal. For the case of arbitrary signal dimension, a set of interpolating functions is described, as well as the procedure for optimizing the adaptive interpolator according to the criterion of the minimum energy of post-interpolation residues. Optimization of the interpolator is carried out for a signal of reduced resolution, after which the found optimal parameter value is used to interpolate the signal of the required resolution. Computational experiments on natural multidimensional signals demonstrate an adaptive interpolator gain of up to 26% as a part of solving the problem of signals fusion.

Andrey Meshcheryakov and Sergei Popov
Using of deep convolutional neural networks for visual features extraction in multiple objects tracking task
The article explores the task of matching visual features when tracing objects on a video sequence. The authors conducted a comparative analysis of existing methods for extracting visual object features and similarity estimation for the re-identification of objects. An algorithm that compares several approaches for feature extraction and similarity estimating has been developed. The authors carried out an experimental assessment of the speed and accuracy of the algorithms using the MOT-16 and PETS 2007 datasets. It is shown that the most accurate estimates of the similarity of objects are achieved by calculating the modified value of the normalized cross-correlation function between features derived from the neural network average pooling layer.

Back to Top