Attestation committee
Accreditation committee
Expert committee
Dispositions, instructions
Normative acts
Scientific councils
Scientific advisers
Doctoral students
Postdoctoral students
CNAA logo

 română | русский | english

CNAA / Theses / 2019 / May /

Analysis and pattern recognition for applications with digital images

Author: Rusu Mariana
Degree:doctor of informatics
Speciality: 01.05.04 - Mathematical modelling, mathematical methods, software
Scientific advisers: Vasile Moraru
doctor, professor, Technical University of Moldova
Horia-Nicolai Teodorescu
academician, doctor, profesor universitar, Universitatea Tehnică Gh.Asachi Iaşi, România
Institution: Technical University of Moldova


The thesis was presented on the 15 May, 2019
Approved by NCAA on the 9 July, 2019


Adobe PDF document0.95 Mb / in romanian


CZU 004.93

Adobe PDF document 6.50 Mb / in romanian
167 pages


image processing, GMM, G-U-MM, forms recognition, correlation, template matching, automatic classification


The thesis is divided into three chapters, followed by bibliography of 188 titles and 4 appendices. The paper contains 60 figures, 25 tables, 116 pages of basic text. The number of published papers on the topic of thesis is 10. Keywords: image processing, GMM, G-U-MM, forms recognition, correlation, template matching, automatic classification. Field of research is image processing in order to pattern/objects recognition and detection in the image. The purpose of this paper is to develop methods, algorithms that would allow the recognition and automatic classification of forms / objects in digital images. The objectives are to analyze, implement and develop image processing algorithms for recognition of forms for digital images applications and make an objective comparison of implemented algorithms. Scientific novelty and originality of the results: there were identified and justified an heuristic segmentation algorithm, a linear separation method for two data sets and a hybrid method of automatic classification of forms / objects. The theoretical importance consists in the development of a histogram-based segmentation method that does not require to indicate the number of thresholds. A mathematical model of linear separation of two sets data has been described. An automatic classification system has been developed that combines different algorithms from artificial intelligence (genetic algorithm, fuzzy system) and defines the particular combination that can provide a better solution for templates / objects classification. The applied value of the thesis: it demonstrated the practical effectiveness of of the proposed segmentation algorithm and the advantage of the hybrid method CCN + AG against the application of individual CCN. Also has been shown effectiveness and advantages of automatic sorting system. The research results can be applied to automatic classification of waste, this area was chosen to emphasize that them correctly sorting would be a solution for reducing pollution. The system can be adapted and developed to classify other objects.