DEPARTMENT SEMINARS
- December, 12.
- PhD graduation with right of defense.
- November, 11.
- The presentation summarizes the research on the contour-based telephone speech Endpoint Detection (ED), which include: - development of new robust features for ED: the FFT magnitude spectrum-based Mean-Delta feature and the Group Delay Mean-Delta feature; - evaluation of the endpoint detection accuracy of the proposed features and two additional ones – the modified Teager energy and the energy-entropy feature; - estimation of the effect of the analyzed ED features in the Dynamic Time Warping fixed-text speaker verification task with short noisy telephone phrases in Bulgarian language.
- January, 17.
- Pre-defense of a dissertation.
- November, 26.
- The following aspects were reviewed and discussed: - The role of low mineralized water solutions for human health; - The influence of acid-alkaline balance on the vital functions of living organisms; - The infuence of the oxidation-reduction potential on the vital functions of living organisms; - Mechanism of activation, methods and means for activating the low mineralized water solutions.
- July, 2.
- The structure of a protein gives more insight in the protein function than its amino acid sequence. Protein structure analysis and comparison is important for understanding the evolutionary relationships among proteins, predicting protein functions, and predicting protein folding. Proteins are formed by two basic regular 3D structural patterns called Secondary Structures (SSs): helices and sheets. A structural motif is a compact 3D protein block referring to a small specific combination of secondary structural elements which appears in a variety of molecules. In this paper we compare a few approaches for motif retrieval based on the Generalized Hough Transform (GHT). A primary technique is to adopt as structural primitives the single SS; alternatives are to adopt as primitive structural element a SSs couple, or a SSs triplet, or up-to an entire motif. The richer is the primitive, the higher is the time for pre-analysis and search, and the simpler the inspection process on the parameter space for analyzing the peaks. Performance comparisons, in terms of precision and computation time, are here presented considering the retrieval of motifs composed of three to five SSs for more than 15 million searches. The approach can be easily applied to the retrieval of greater blocks, up to protein domains, or even entire proteins.
- June, 25.
- The prezentation includes information about the investigations carried out by scientists from different countries during the last decades on the anomaly properties of the electro-chemically treated water. The way of producing activated water, its properties and action are described. Experimental results obtained in medical and scientific institutions in Russia, Japan, South Africa and some Bulgarian investigators are shown. The broad possibilities for the application of activated water in health care, agriculture and ecology are underlined.
- November, 13.
- We look upon a practical approach for emotion (anger, disgust, fear, happiness, sadness and surprise) recognition from video clips based on the face dynamics, and supporting experiments. Recognition is carried out in three distinctive steps: 1) localization of landmarks, where an automatic method as well as the result of annotated frames are used; 2) the superposition of a three dimensional wireframe over the video frames using an iterative method for solving a linear multivariate system equation consisting of the wireframe parameters – position, size and face dynamics and the comparison of the transformed wireframe with the landmarks from the previous step by the least squares method; 3) classification of the face dynamics parameters from the previous step using Fisher's linear discriminant and separating an emotion from the subset of all the other emotions.
- June, 28.
- This presentation concerns the work done during the PhD period. The goal and the tasks of the dissertation titled “Combined approach to signature recognition” are being presented. In order to solve the corresponding classification task, the following steps are being performed: signature data collection, feature extraction, feature selection and classification. GUI interface developed to facilitate the process and the database diagrams are also presented. The summary of the future work is also described.
- April, 20.
- An advanced radar system, developed by the JRC, is being used at the Giglio Island in Italy to support the Italian authorities in monitoring the movements of the cruise ship that grounded off Tuscany on 13th January. Data on the ship movements have been of fundamental importance during the search and rescue operations and will remain crucial when the fuel is pumped off the ship's tanks. The JRC system, called MELISSA, is an innovative and low-cost radar imaging system that allows for the detection of movements of objects with very high accuracy and resolution. The system is compact, portable and easily deployable and can work day and night under any weather conditions.
- March, 27.
- On 13/01/2012 at Giglio island a 400 long ship “Costa Concordia” having about 4000 people on board shipwrecked. Since then, it has been constantly moving downward the sea bottom. Additionally, the unusual ship position, the weather conditions and the rescue operations impose on the construction additional forces, not taken into consideration in its engineering design, which may consequently jeopardize the rescue teams and the environment. Aiming maximal safety of the rescue operations, continuous monitoring of the ship displacement has been launched. This tragedy results to analysis about the preparation and the possibility of adequate crisis reactions. The presentation reveals the ship displacement monitoring system, putting into practice MIMO radar interferometer – a scientific work of the Joint Research Centre - Italy and personal impressions of the ship rescue activities.
- January, 24.
- November, 3.
- In this paper, we propose a method that, in contrast to traditional classification methods based on Hyper Surface Classification (HSC), uses two - one inner and one outer - classification hyper-surfaces (CHS). The outer CHS includes the inner CHS together with the set that it delimits. These two CHS allow the authors to define a new classification rule that uses the smallest distance between the point to be classified and the intersection points of the double CHS with a ray starting at this point. In contrast to known classification rules used in HSC-methods, this innovative rule is reliable and easy to implement in practice, which makes the proposed classification method accurate and efficient.
- September, 7.
- The recent immersion of depth/range cameras triggered new wave of research in the area of human pose recognition. Depth images provide grounds for a new feature space that can be used to build strong and robust classifiers. Nowadays there exist open source frameworks providing algorithms for real time human parts recognition and tracking. But the problem of recognizing computer commands described by human gestures is still very specific to the application area. More generalized approach for defining descriptive language is required to allow wider usage of natural user interfaces (NUI), where the systems are controlled with native human gestures. The presentation proposes a method for mapping a gesture to a handwritten curve and an attempt to recognize it via Dynamic Time Warping algorithm.
- The complex human body part segmentation task is interpreted as an object classification problem. The Random Forest (RF) method is used to infer a set of probabilities for each data pixel, each one indicating the probability of a pixel to belong to a particular label - body part. The analysis involves depth map images as they are independent on the scene configuration, e.g. illumination, texture, viewpoint. The somewhat fuzzy boundaries between the RF segmented parts are defined more precisely through the use of the Graph Cut method.
- May, 25.
- Each education system needs to be adequate to the contemporary social development. In the last decades the traditional methods in the educational processes exhibit more and more limitations. This lets new methods and technologies, in accordance to the scientific and technological development, make their way in the education system. The constant progress in the information and communication technologies leads to the need of a distant form of education. The goal of this paper is to make an analysis of the needs and to demonstrate the necessity of the virtual education environments.
- This paper examines a simple method for determining the distance to objects in a scene by stereo-images, using the principles of a canonical stereovision system. This method is based on finding corners (characteristic points) in each image and then finding the correspondence between these corners. The distances to different objects are calculated by horizontal shifting's length between these correspondences. The experiments are carried out over a static structured scene, obtained by a single camera, which is horizontally translated in different distances, thus imitating a canonical stereovision system.
- May, 16.
- In this paper, we approach feature set selection phase in signature verification by applying the method for selection of regression variables based on Mallows Cp criterion for regression. In this way we identify best feature subsets of various sizes for each user of our database on the basis of his/her ten genuine and ten random forgery on-line signatures. Among these subsets we select the best subset that have Cp value closest to p, where p is the number of regression coefficients. Thus, we obtain for each user the best feature subset of a different size. Our aim is to check whether there are common features among best feature subsets for all users which will justify the removal of the rest features from the initial feature set. The results obtained with the database of 140 signatures collected from fourteen users demonstrated that we cannot restrict to a common feature set valid for all users but instead of that we have to consider each user best feature set separately in signature verification.
Department SPPR