Vol 26, No 3 (2016)

The problem of choosing an ensemble of binary subproblems for multiclass recognition problems is considered in the algebraic and logical approaches. The connection of the first and second level algorithm correctness and the applicability of logical deduction in a multilevel scheme are studied. Two modifications of the ECOC method are proposed that allow one to improve the quality of the original method under different conditions; this is demonstrated using model and applied problems.

In order to obtain tractable formal descriptions of poorly formalized problems within the context of the algebraic approach to pattern recognition, we develop methods for analyzing metric configurations. In this paper, using the concepts of σ-isomorphism and σ-completion of metric configurations, a system of criteria for assessing the properties of “generalized density” is obtained. The analysis of the density properties along the axes of a metric configuration allowed us to formulate methods for calculating the topological neighborhoods of points and for finding the “grains” of metric condensations. The theoretical results point to a new plethora of algorithms for searching metric condensations − methods based on the “restoration” of the set (the condensation searched) using the data on the components of the projection of the set on the axes of the metric configuration. The only mandatory parameters of any algorithm of this family of algorithms are the metric itself and the distribution of σ, which characterizes the accuracy of the values of the metric.

In consideration of living organisms’ ability to endure for years, and their ability to adapt to surrounding environment, the mechanism of evolution is the inspiration for creating a new genetic algorithm. The goal of this paper is to examine possibilities of genetic algorithm application for segmentation of digital image data, implementation of this algorithm, and to create tools for its testing. The next goal is to examine possible choices of algorithm’s parameters, and to compare quality of the results with other segmentation methods within various image data.

The paper is devoted to handling wideband monitoring tasks by discrete Fourier transform (DFT) modulated filter banks. Filter bank implementation is considered using CPU (Central Processing Unit) and CUDA (Compute Unified Device Architecture) based on GPUs (Graphics Processing Units). We show that CUDA is more efficient for big signal sets due to low temporal and computational costs. The paper also discusses signal classification in filter bank channels for different signal-to-noise ratios using binary decision trees (with the iterative Adaboost procedure) and neural networks. The total classification error in our experiments does not exceed 10%. The results can be extended and applied to hydroacoustic monitoring tasks.

This paper is devoted to construction of a hybrid intelligent system of express-diagnostics of possible information security attackers (HIS DIVNAR) based on a synergy of several sciences and scientific directions: test pattern recognition; discrete mathematics; threshold and fuzzy logic; artificial intelligence; finite state machines (FSM) theory; reliability; theory of separating systems; theory of probability and mathematical statistics; and cognitive means. The proposed approach and basis of the mathematical apparatus are fragmentarily given for constructing HIS DIVNAR; that consists of four components: the first component, called IS DIOS, is designed for the express-diagnostics of organizational stress of the subject; the second (IS DIAPROD) is for the express-diagnostics and prevention of depression; the third (DIDEV) is for the expressdiagnostics and prevention of deviant behavior; and the fourth, intelligent system of express-diagnostics of information security attackers (IS DINARLOG2) is for making and justification of decisions with the use of cognitive means based on earlier revealed different regularities, including fault-tolerant irredundant unconditional diagnostic tests, fault-tolerant mixed diagnostic tests, regularities, and decision rules, which are built by using the applied IS DINARLOG1 constructed on the basis of intelligent instrumental software IMSLOG. Further development of this approach is proposed.

This paper presents the results of development of a novel method for measuring nonstationary quasi-periodic biomedical signals, in particular, the arterial blood pressure pulse signal. It has been demonstrated that the proposed method for compensation tracing of dynamic signals suggests not only smart, but also active sensors. In connection with this, a major part of the introduction is devoted to expanding the conception of smart sensors to the paradigm of active sensors. Further, following the introduction on the background of the question, a brief description of the functioning principles and some design features of the active sensor developed by us are given. The results of the sensor test and calibration are discussed, and the necessity of its complicated control is substantiated. The remaining part of the paper is devoted to possible ways of development of this control and the way that we have chosen to control the active sensor of arterial blood pressure. The principle of controlling compensation of a pulse pressure based on prediction of pulse wave patterns is discussed and substantiated. The final part is devoted to technical matters of formation of dynamic patterns using multiscale correlation analysis of a current local period of heart contractions.

The paper presents an algorithm for quantification the degree of receptor expression to steroid hormones by automatic analysis of microscope images of immunocytochemical specimens. During experiments a high correlation between the results of the automatic analysis and visual expert assessment was shown and the possibility to apply the proposed algorithm to automate immunocytochemical analysis was confirmed.

In order to detect the traversable region of automotive land vehicle (ALV), a multi-scale data analysis and representation method, shearlet transform is researched. Based on the feature that the descriptor calls Histograms of Shearlet Coefficients (HSC), a weighted HSC (WHSC) is proposed. Compared to HSC, WHSC uses the scale factor, which makes it better than HSC. We combine WHSC and color histogram in HSV color space as the fusion-feature, and use partial least squares (PLS) to project the high dimensional feature vectors onto a subspace. Also, support vector machine (SVM) is used based on linear kernel as the classification to solve traversable region detection. The experiment results suggest that for both in NUSTrobot dataset and OUTEX dataset, the method provided by this paper performs much better, and can detect the traversable regions in complex environments (e.g., different shadow and lighting conditions). Moreover, with the help of this method, the platform can achieve more functions.

Facial expression recognition is a challenging field in numerous researches, and impacts important applications in many areas such as human-computer interaction and data-driven animation, etc. Therefore, this paper proposes a facial expression recognition system using active shape model (ASM) landmark information and appearance-based classification algorithm, i.e., embedded hidden Markov model (EHMM). First, we use ASM landmark information for facial image normalization and weight factors of probability resulted from EHMM. The weight factor is calculated through investigating Kullback-Leibler (KL) divergence of best feature with high discrimination power. Next, we introduce the appearance-based recognition algorithm for classification of emotion states. Here, appearance-based recognition means the EHMM algorithm using two-dimensional discrete cosine transform (2D-DCT) feature vector. The performance evaluation of proposed method was performed with the CK facial expression database and the JAFFE database. As a result, the method using ASM information showed performance improvements of 6.5 and 2.5% compared to previous method using ASM-based face alignment for CK database and JAFFE database, respectively.

A language model based on features extracted from a recurrent neural network language model and semantic embedding of the left context of the current word based on probabilistic semantic analysis (PLSA) is developed. To calculate such embedding, the context is considered as a document. The effect of vanishing gradients in a recurrent neural network is reduced by this method. The experiment has shown that adding topic-based features reduces perplexity by 10%.

Content Based Image Retrieval (CBIR) systems use Relevance Feedback (RF) in order to improve the retrieval accuracy. Research focus has been shifted from designing sophisticated low-level feature extraction algorithms to reducing the “semantic gap” between the visual features and the richness of human semantics. In this paper, a novel system is proposed to enhance the gain of long-term relevance feedback. In the proposed system, the general CBIR involves two steps—ABC based training and image retrieval. First, the images other than the query image are pre-processed using median filter and gray scale transformation for removal of noise and resizing. Secondly, the features such as Color, Texture and shape of the image are extracted using Gabor Filter, Gray Level Co-occurrence Matrix and Hu-Moment shape feature techniques and also extract the static features like mean and standard deviation. The extracted features are clustered using k-means algorithm and each cluster are trained using ANN based ABC technique. A method using artificial bee colony (ABC) based artificial neural network (ANN) to update the weights assigned to features by accumulating the knowledge obtained from the user over iterations. Eventually, the comparative analysis performed using the commonly used methods namely precision and recall were clearly shown that the proposed system is suitable for the better CBIR and it can reduce the semantic gap than the conventional systems.

Anti-aging and looking young with a full of vigor appearance with no Facial volume depletion and deepening lines of facial expression is a dream of every human being in life. Researchers in dermal and cosmetic fields had spent many years looking for solutions to aging signs and wrinkles other than surgeries. Botox is a skin rejuvenation cosmetic procedure that represents the recent magical key to aging appearance problems especially with the fascinating results it had showed. Botox can simply make you look 10 to 20 years younger, which represent an obstacle in the face of human age estimation researches. In this paper, we proposed a new model called Human Injected by Botox Age Estimation (HIBAE) model, a human age estimator based on active shape models, speed up robust feature, and support vector machine to accurately estimate the age of people that are exposed to Botox injections. Human Injected by Botox Age Estimation proposed model was trained by a crossover of Productive Aging Lab. database and 60 images collected from the internet of people that were exposed to Botox, and tested using a crossover of FACES64 database and 20 images of people that were exposed to Botox. HIBAE had showed superiority through performance testing over the state-of-the-art.

This paper addresses the problem of automating the production of digital graphic documents based on pattern recognition and machine learning methods. Methods for solving problems of detecting and isolating objects with similar descriptions, as well as their parallelization, are considered. Interactive procedures of editing and assigning classification codes to the etalons constructed are described. The results are demonstrated with an example of inputting real documents.

Signatures of medical contouring of the left-ventricular (LV) regions are studied. This research is part of addressing the general problem on automatic contouring of LV regions on ultrasound frames with apical four-chamber cardiac projection. A signature is a curve of an LV contour plotted in polar coordinates. The optimal point in the center of the LV contour base has been identified. The resulting signature was approximated by three second- and third-degree polynomials, i.e., the left side, the right side, and the vertex. The result of this approximation was qualitatively and quantitatively better than that of the entire curve.