Papers by Dejan Gjorgjevikj
Citeseer
... nocnen ~ ~ ~ e 8 o6enexja ce ~ O ~ H ~ HM On pa3JIHsHH perHOHEI ( 3 0 ~ ~ ) Ha CJIkIKaTa KaKO... more ... nocnen ~ ~ ~ e 8 o6enexja ce ~ O ~ H ~ HM On pa3JIHsHH perHOHEI ( 3 0 ~ ~ ) Ha CJIkIKaTa KaKO IIITO e npliKamaHo Ha Cn. 3. O ~ oj Tkin Ha o6enemja BO pa3n~q~a @op~a ceKO-PHCTeHH BO MHOrY CHCTeMki 38 IIpen03HaBaHS CHM~OJIM ( ~ a npaMep [lo]). ...
SPIE Proceedings, 1998
Shot boundary detection is fundamental to video analysis since it segments a video into its basic... more Shot boundary detection is fundamental to video analysis since it segments a video into its basic components. This paper presents a comparison of several shot boundary detection techniques and their variations including color histogram, edge directions histogram and wavelet transformations statistics. The performance and ease of selecting good thresholds for these algorithms are evaluated based on a wide variety of video sequences with different object and camera motions. Threshold selection is performed using sliding window. We used TV news, sports and documentary, music, movie and nature video sequences to estimate the performance of the algorithms. The experimental results indicate that the algorithm based on color histograms is most suitable for shot boundary detection in film and documentary categories, but the algorithm based on wavelet is preferable for nature and sports categories.
Beekeeping practice, being very environmentally dependent, requires the temperature and the humid... more Beekeeping practice, being very environmentally dependent, requires the temperature and the humidity in the hive to be in some regular ranges for optimal beehive health and productivity. Since most of the plants and flowers required for beehive prosperity and honey production are usually outside inhabited areas, the beekeeper must travel to the bee colonies to check them, which can be time and resource consuming. In this paper, an end to end remote monitoring and control system for a bee colony is presented. The system is consisted of a web-based system for monitoring and control of the conditions of the hives and IoT system for collecting the sensor measurements and transferring the data. The IoT system is composed of hardware units that are mounted on the beehives, containing temperature, humidity, weight sensors, actuators, and a microcontroller responsible for collecting the measurements and sending the data to the web system. The communication between the hardware unit and the web system uses WiFi or LoraWAN technology, that enables running the device on batteries. The system enables remote monitoring of multiple beehives and can be configured to alert the user via email or push notification if some sensor value is outside of predefined range. The system also enables sending commands to the unit controlling the actuators that can intervene on the beehive closing or opening a ventilation lid.
Disorder Aleksandar Teneva, Slobodan Kalajdziskia, Dejan Gjorgjevikja, Ljupco Kocarevb, Eduard Vi... more Disorder Aleksandar Teneva, Slobodan Kalajdziskia, Dejan Gjorgjevikja, Ljupco Kocarevb, Eduard Vietac*, Dina Popovicc**, Luis Pintorc, Pablo Villosladac, Vesna Prckovskac a Faculty for Computer Science and Engineering, Sts. Cyril and Methodius University, Skopje, R.Macedonia b Macedonian Academy of Sciences and Arts, Skopje, R. Macedonia c Center for Neuroimmunology, Department of Neurosciences, IDIBAPS, Hospital Clinic, Barcelona, Spain c* Clinical Institute of Neuroscience, University Hospital Clinic of Barcelona, Barcelona, Spain. C** Bipolar Disorders Program, Institute of Neurosciences, University Hospital Clinic of Barcelona, Barcelona, Spain.
This paper addresses the problem of combining the results of multiple adaptive logic networks tha... more This paper addresses the problem of combining the results of multiple adaptive logic networks that were used as classifiers in OCR system for Macedonian Cyrilli c. Combining is performed by weighted voting of some or all the classifiers. The weights associated to every trained classifier are evaluated on a test sample. As a result of the weighted voting process, a li st of character candidates along with their confidence levels is produced for every pattern. The confidence levels of the character candidates are used in the contextual postprocessing phase for word candidates generation and lexicon lookup.
Advances in Intelligent Systems and Computing, 2016
This work presents an approach for blocking artifacts removal in highly compressed video sequence... more This work presents an approach for blocking artifacts removal in highly compressed video sequences using an algorithm based on dictionary learning methods. In this approach only the information from the frame content is used, without any additional information from the coded bit-stream. The proposed algorithm adapts the dictionary to the spatial activity in the image, by that avoiding unnecessary blurring of regions of the image containing high spatial frequencies. The algorithms effectiveness is demonstrated using compressed video with fixed block size of 8x8 pixels.
Proceedings of the 23rd International Conference on Information Technology Interfaces, 2001. ITI 2001., 2001
Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004., 2004
EUROCON 2005 - The International Conference on "Computer as a Tool", 2005
11th IEEE Mediterranean Electrotechnical Conference (IEEE Cat. No.02CH37379)
MELECON '98. 9th Mediterranean Electrotechnical Conference. Proceedings (Cat. No.98CH36056)
... filtering and thresholding, skew detection and possibly correction, and segmentation. In most... more ... filtering and thresholding, skew detection and possibly correction, and segmentation. In most existing OCR systems, character recognition is performed on individual characters. Therefore, prior to character recognition it is necessary to isolate individual characters froin the text ...
A subsystem for text-to-speech (TTS) conversion for Macedonian language as a part of a system for... more A subsystem for text-to-speech (TTS) conversion for Macedonian language as a part of a system for support of humans with damaged sight will be presented in this paper. The whole system includes recognition of printed Cyrillic text, archiving, TTS conversion and printing on a Braille printer. None of tested commercial programs for generating human speech on personal computer is appropriate for Macedonian language. Therefore a subsystem for real-time TTS conversion from unrestricted text is under development. The whole architecture of the subsystem and some of it's major modules will be presented here. Number of techniques, like automatic generation of prosodic rules, time scale modification of speech based on nonlinear oscillator model and flexible speech units inventory will be used to overcome the fact that Macedonian language is comparatively poorly researched
ABSTRACT In this paper, various cooperation schemes of SVM (Support Vector Machine) classifiers a... more ABSTRACT In this paper, various cooperation schemes of SVM (Support Vector Machine) classifiers applied on two feature sets for handwritten digit recognition are examined. We start with a feature set composed of structural and statistical features and corresponding SVM classifier applied on the complete feature set. Later, we investigate the various partitions of the feature set as well as the advantages and weaknesses of various decision fusion schemes applied on SVM classifiers designed for partitioned feature sets. The obtained results show that it is difficult to exceed the recognition rate of a single SVM classifier applied straightforwardly on the complete feature set. Additionally, we show that the partitioning of the feature set according to feature nature (structural and statistical features) is not always the best way for designing classifier cooperation schemes. These results impose need of special feature selection procedures for optimal partitioning of the feature set for classifier cooperation schemes.
Abstract: Despite of the presence of many commercial OCR programs on the market, therecognition a... more Abstract: Despite of the presence of many commercial OCR programs on the market, therecognition accuracy is still a difficult problem, especially when handling multilingualdocuments. In absence of an OCR program which can concurrently recognize mixed scriptdocuments, we have developed an OCR program for recognizing mainly Macedonian text withwords in Latin script. In this paper an adaptation of preclassification based on character shape,for Macedonian characters along with standard Latin characters is presented. Preclassificationis performed using the line parameters such as baseline and upper-baseline, as well as thedimensions of the character boxes and their horizontal position within the text line. Keywords: OCR, character preclassification, mixed script text recognition Although speech is a sign system that is more natural than writing, writing is considered to havemade possible much of the culture and the civilization. Different writing systems, or scripts,represent linguistic units, words, syllables and phonemes, at different structural levels. Somelanguages ignoring minor differences in orthography share the Latin script, and some alsoignoring minor differences in orthography share the Cyrillic script. Each script has its own set oficons, known as characters or letters, that have certain basic shapes. In addition to linguisticsymbols, each script has a representation for numerals, and some special symbols.The task of text recognition is to transform the image data containing textual information into its"usual" representation as a strings over a given alphabet (e.g. ASCII). Today's commercial OCRsystems reach highly reliable results on good quality documents. However, in the case ofmultiple scripts, current devices are far from recognition accuracies demanded for subsequentautomatic text analysis. Systems that recognizes characters from a scanned image can be dividedinto three main operational steps: document layout analysis, character recognition, andcontextual postprocessing.Document layout analysis consists of image filtering and thresholding, skew detection andpossibly correction, and character segmentation. In most existing OCR systems, characterrecognition is performed on individual characters. Therefore, prior to character recognition it isnecessary to isolate individual characters from the text image. The technique which partitionsthe image of lines and words in individual characters is called character segmentation.Approaches for segmenting document image components can be divided in top-down andbottom-up approaches. Top-down techniques divide the region into major regions which arefurther divided into subregions, while bottom-up methods first extract small elements (typically
Object recognition supported by user interaction for service robots
Measurement Science and Technology, 2021
Different machine learning approaches have been developed for the fault diagnosis of mechanical s... more Different machine learning approaches have been developed for the fault diagnosis of mechanical systems. To achieve desired diagnosis performance, lots of labeled one-dimensional (1D) signals are required for training machine learning models. however, those signals collected under various working conditions are difficult to be used for both diagnosis model training and testing. For real applications, moreover, the collection of labeled data is more difficult than that of unlabeled ones. To tackle the above challenging points, a dynamic transfer adversarial learning (DTAL) network is proposed for dealing with unsupervised fault diagnosis missions. To this end, an improved feature extractor is developed to deal with 1D mechanical vibration signals. A dynamic adversarial factor is presented to automatically adapt the marginal distribution of the global domain. The conditional distribution of the local domain is employed to make the model independent of training multiple classifiers, so as to reduce the computational burden of the proposed method. The addressed DTAL was evaluated using fault diagnosis experiments for a wind turbine gearbox and benchmark bearings. Compared with other state-of-the-art methods, it has better accuracy and robustness as highlighted by experimental results. The developed model can improve the diagnosis performance under various workloads for mechanical systems.
Novelty detection is a challenging task for the machinery fault diagnosis. A novel fault diagnost... more Novelty detection is a challenging task for the machinery fault diagnosis. A novel fault diagnostic method is developed for dealing with not only diagnosing the known type of defect, but also detecting novelties, i.e. the occurrence of new types of defects which have never been recorded. To this end, a sparse autoencoder-based multi-head Deep Neural Network (DNN) is presented to jointly learn a shared encoding representation for both unsupervised reconstruction and supervised classification of the monitoring data. The detection of novelties is based on the reconstruction error. Moreover, the computational burden is reduced by directly training the multi-head DNN with rectified linear unit activation function, instead of performing the pre-training and fine-tuning phases required for classical DNNs. The addressed method is applied to a benchmark bearing case study and to experimental data acquired from a delta 3D printer. The results show that it is able to accurately diagnose known ...
This paper presents architecture of Support Vector Machine classifiers arranged in a binary tree ... more This paper presents architecture of Support Vector Machine classifiers arranged in a binary tree structure for solving multi-class classification problems with increased efficiency. The proposed SVM based Binary Tree Architecture (SVM-BTA) takes advantage of both the efficient computation of the tree architecture and the high classification accuracy of SVMs. Clustering algorithm is used to convert the multi-class problem into binary tree, in which the binary decisions are made by the SVMs. The proposed clustering model utilizes distance measures at the kernel space, not at the input space. The performance of this method was measured on the problem of recognition of handwritten digits and letters using samples from MNIST, Pendigit, Optdigit and Statlog database of segmented digits and letters. The results of the experiments indicate that this method has much faster training and testing times than the widely used multi-class SVM methods like "one-against-one" and "one-against-all" while keeping comparable recognition rates. The experiments showed that this method becomes more favorable as the number of classes in the recognition problem increases.
Uploads
Papers by Dejan Gjorgjevikj