Vol 28, No 4 (2018)

The Histogram of Oriented Gradients (HOG) is one of the most widely used methods to extract the gradient features for object recognition and consistently shows high accuracy rates when compared to other descriptors. The major drawbacks of using the HOG method are the necessity of finding an optimal window size to fit the whole object; and the exhaustive search mechanism represented by a fixed window size sliding through the whole image to locate and recognize objects. This research proposes an efficient and robust Dynamic-HOG as an improvement to the traditional HOG method to locate and recognize structured objects in images. The proposed method works by locating and analyzing the structured objects in images in order to define a dynamic window size w.r.t. each object size. Moreover, the Dynamic-HOG method requires much less processing time by eliminating the exhaustive search mechanism. The method defines the height and width thresholds of objects and bounds each object with a window w.r.t. its size while ignoring non–object edges. It fits structured objects of a close range of heights and widths. This paper considers the characters that are minted on coins of different languages and sizes as the objects to recognize. There are several papers in the literature discussing coin recognition problem and proposing solutions based on various sets of features extracted from the entire coin image. This research also proposes a new method for coin recognition by focusing on recognize coins based on smaller part of the coin image which are the characters. Our method is evaluated on coins from diverse countries with different background complexity. The proposed method achieved precision and recall rates as high as 98.08 and 98.23%, respectively; which demonstrate the effectiveness and robustness of the proposed method.

Recently, raw experimental data in machine learning often appear as direct comparisons between objects (featureless data). Different ways to evaluate difference or similarity of a pair of objects in image and data mining, image analysis, bioinformatics, etc., are usually used in practice. Nevertheless, such comparisons often are not distances or correlations (scalar products) like a correct function defined on a limited set of elements in machine learning. This problem is denoted as metric violations in ill-posed matrices. Therefore, it needs to recover violated metrics and provide optimal conditionality of corresponding matrices of pairwise comparisons for distances and similarities. This is the correct basis for using of modern machine learning algorithms.

The paper summarizes the joint work of specialists in the fields of image analysis and ophthalmology over the last few years. As a result of this work, new mathematical methods for automating image analysis that have important diagnostic value for ophthalmology have been developed: (1) identification of the lipid layer state in the intermarginal space of human eyelids; (2) analysis of the degree of cellular structure density (cellularity) in the corneal tissue of human eyes; (3) identification of the state of the retinal blood flow when analyzing fluorescent angiograms of the human fundus; and (4) morphometric analysis of the state of the epithelium posterius (endothelium) in the human eye cornea. As initial data, we used (respectively) (1) images of imprints of the eyelid intermarginal space on a millipore filter upon their osmium vapor staining; (2) micrographs of corneal tissue specimens obtained using a light microscope; (3) fluorescent angiograms of the human fundus; and (4) images of endothelial cells obtained noninvasively using a confocal microscope. The developed methods are designed to extract morphometric data from these images. For each problem, the following results were obtained: (1) expectations and variances of pixel intensities on the imprint along a drawn line and over a selected region, as well as plots that characterize pixel intensity and change in the thickness of the imprint along a drawn line; (2) expectations and variances for the intensities of the selected regions and intensity histograms; (3) extracted vessels and ischemia zones with their statistical descriptions; and (4) detected cells of hexagonal, pentagonal, and other shapes, as well as a set of characteristics associated with the size of the cells detected. The developed methods are based on the fundamental results of the mathematical theory of image analysis and on the joint use of image processing, mathematical morphology, and mathematical statistics techniques. The paper also describes software implementations of the developed methods, including an automated research workstation for ophthalmologists, and presents the results of their experimental testing.

Photoplethysmography (PPG) is a method for contactless heart rate estimation through the analysis of slight variations of skin color. Skin color variation caused by changes in the blood volume in vessels and registered by a camera is called color signal. Recent studies proved that some PPG methods could be used to produce accurate heart rate estimations on videodata recorded by common web-cameras that makes them potentially applicable for longterm health monitoring in home or office conditions. In this work, we study novel Approximation-based transformation method of signal processing and evaluate its combination with common preprocessing and postprocessing algorithms. Approximation-based transformation is the procedure of computing an approximation signal that consists of leading coefficients of the local quadratic approximation of the color signal.

Fine-grained classification is a recognition task where subtle differences distinguish between different classes. To tackle this classification problem, part-based classification methods are mostly used. Partbased methods learn an algorithm to detect parts of the observed object and extract local part features for the detected part regions. In this paper we show that not all extracted part features are always useful for the classification. Furthermore, given a part selection algorithm that actively selects parts for the classification we estimate the upper bound for the fine-grained recognition performance. This upper bound lies way above the current state-of-the-art recognition performances which shows the need for such an active part selection method. Though we do not present such an active part selection algorithm in this work, we propose a novel method that is required by active part selection and enables sequential part-based classification. This method uses a support vector machine (SVM) ensemble and allows to classify an image based on arbitrary number of part features. Additionally, the training time of our method does not increase with the amount of possible part features. This fact allows to extend the SVM ensemble with an active part selection component that operates on a large amount of part feature proposals without suffering from increasing training time.

The ability to provide reliable protection against counterfeiting is one of the key requirements for a biometric security system. Iris recognition as the technology emerging on mobile market is assumed to handle various types of spoof attacks to prevent compromise of the user’s personal data. A method of iris anti-spoofing is proposed in this work. It is based on applying of convolutional neural network and capable to work in real-time on the mobile device with highly limited computational resources. Classification of iris sample for spoof and live is made by a single frame using a pair of images: eye region and normalized iris. The following types of iris spoof samples are considered in this particular work: printed on paper, printed on paper with imposition of a contact lens, printed on paper with application of transparent glue. Testing of the method is performed on the dataset manually collected and containing all the mentioned spoof sample types. The method revealed its high performance in both classification accuracy and processing speed as well as robustness under uncontrollably changing environmental conditions, which are specific and significant when interacting with the mobile device.

The problem of optimally choosing a learning object is studied. The definition of essential totality of precedents is given. It is shown that an iterative procedure for generating the essential totality of precedents exists. The algorithm operation is checked on model data.

We analyze the cluster structure of learning samples, decomposing class objects into disjoint groups. Decomposition results are used for the computation of the compactness measure for the sample and its minimal coverage by standard objects. We show that the number of standard objects depends on the metric choice, the distance to noise objects, the scales of the feature measurements, and nonlinear transformations of the feature space. We experimentally prove that the set of standards of the minimal coverage and noise objects affect the algorithm generalizing ability.

The problem of data visualization in the analysis of two classes in a multidimensional feature space is considered. The two orthogonal axes by which the classes are maximally separated from each other are found in the mapping of classes as a result of linear transformation of coordinates. The proximity of the classes is estimated based on the minimum-distance criterion between their convex hulls. This criterion makes it possible to show cases of full class separability and random outliers. A support vector machine is used to obtain orthogonal vectors of the reduced space. This method ensures the obtaining of the weight vector that determines the minimum distance between the convex hulls of classes for linearly separable classes. Algorithms with reduction, contraction, and offset of convex hulls are used for intersecting classes. Experimental studies are devoted to the application of the considered visualization methods to biomedical data analysis.

This paper discusses the possibility of using the classical results of graph theory related to reconstruction and recognition of graphs and their characteristics in the field of image recognition. Two heuristic approaches are proposed to estimate the adequacy of object images. Various aspects of the problem of graph description (representation) with the use of graph invariants are analyzed. New classes of invariants that can be used to construct the heuristics mentioned above are introduced and investigated. In addition, some statements concerning two aspects of the problem—the formation of complex invariants taking into account the basic and functional dependences among invariants—are proved.

In last years, several biometric modalities are coming back into the field of education. Especially, face recognition modality that allows authenticating humans based on their facial features. The main objective of this paper is to implement the recognition system part for a global software that is intended to be used in many applications related to the educational field. Therefore, we propose to combine the face description method based on Local Gradient Probabilistic Pattern (LGPP), the two dimensional subspace methods, and machine learning classifiers. Firstly, we extract principal face component using the LGPP descriptor. Then, 2DDWT, 2DPCA, and 2DLDA subspace methods were implemented to reduce face features. Finally, support vector machine (SVM) and artificial neural network (ANN) based machine learning algorithms are applied to classify the set of the final features vectors. The experimental results on relevant databases demonstrate the effectiveness of the proposed system.

The paper presents the design and implementation of software tools for statistical analysis of the real data based on the assumptions that empirical distribution can be approximated by generalized negative binomial (GNB) or generalized gamma (GG) families. Models based on GG distributions are widely applied in such practical problems as processing of synthetic-aperture radar images and speech signals, hydrological analysis and optical communications. In this paper, the GNB distributions are considered as a mixed Poisson law with the mixing GG distribution. This family could provide better fit with the different statistical data than classical negative binomial distributions that have been successfully used for analysis of precipitation events earlier. The parameter estimation is implemented using a functional approach, so approximations by different types of distributions are compared in sense of different metrics. The results of application of the implemented software tools are demonstrated on the example of the Potsdam precipitation events.

Traditional wildfire detection methods are of low efficiency and cannot meet user needs, a novel method based on deep time and space features along transmission line is proposed in this paper, which uses ViBe algorithm to detect movements in videos, and extracts static deep feature in the space domain and dynamic optical flow feature in the time domain respectively. At last the deep convolutional neural network model in cascade is used to classify and find out real wildfire regions. By using combined deep features extracted from dynamic time-domain and static space-domain respectively, our method can eliminate the interference of movements of other objects with similar colors.

The testing of the proposed method for estimating the geometric parameters of a three-dimensional object from the data of computer-aided simulation and field experiments confirmed by excavation provide the possibility to determine the location of anomalous-resistance objects under the ground and to perform the quantitative estimation of their geometric parameters (shape, dimensions, and burial depth) from resistivity survey data. It is shown that the analysis of the vector pictures of main resistance change directions provides the possibility to estimate the spatial location of an object of search (along the depth and in horizontal “sections”) at a qualitative level and to determine its relative resistance. The application of the scalar product function and the adaptive fuzzy clustering algorithm to these vector pictures provides the possibility to estimate the shape of an object of search and the range of its burial depths. Using the A* optimal path search algorithm, it is possible to plot the boundary line of an object on a set of horizontal “sections.”