TSI logo 25 years
LV EN
Home > About Us > Structure and Government > Abstracts

DEGREE

PROGRAMME

FACULTY

YEAR

LANGUAGE

Improving the accuracy of optical character recognition of stone engravings using image pre-processing methods

This study focuses on the development of preprocessing methods to improve the accuracy of Optical Character Recognition (OCR) for stone engravings. The primary goal is to enhance the precision of widely used OCR tools, particularly for texts engraved on stone surfaces, which present unique challenges that differ from traditional OCR applications. Emphasis is placed on developing image preprocessing methods as a software product. Customized image manipulation scripts were used to improve recognition accuracy and address issues such as contrast, alignment, noise, and resolution. The preprocessing stage was integrated into the workflow designed for image transformation before OCR processing. Subsequently, the recognition improvements were evaluated based on text similarity metrics analysis. Iterative text recognition and repeated recognition of images after applying preprocessing demonstrated significant improvements in OCR accuracy. This work provides a solid foundation for further enhancement of OCR workflows by employing adaptable preprocessing techniques specifically designed for particular problem areas, achieving higher precision in text recognition.

Author: Romans Urbans-Orbans

Supervisor: Aleksandrs Grakovskis

Degree: Bachelor

Year: 2024

Work Language: Latvian

Study programme: Computer Science

More...

Enhancement Strategies for Retrieval-Augmented Generation Systems

This thesis systematically explores the enhancement of Retrieval-Augmented Generation (RAG) systems within Large Language Models, emphasizing optimization of retrieval parameters and generation accuracy. We investigate optimal configurations in RAG systems, including chunk size and overlap percentages, top-k selection, query transformations, different retrieval methods, different LLMs, namely GPT-3.5-Turbo and GPT-4, discovering that a chunk size of 500 tokens generally offers the best performance. Vector search using cosine similarity emerges as the most effective retrieval method, significantly enhancing both context precision and recall across various tasks and knowledge bases. Experimentation within the CRUD-RAG framework demonstrates its applicability in diverse tasks from content creation to knowledge refinement. Our findings indicate that enhancements in retrieval settings can markedly improve the performance of RAG systems, making them more efficient and adaptable for complex information synthesis and retrieval tasks. These results affirm the potential of systematic enhancements to improve AI-driven language models in practical applications, contributing significant insights and practical approaches to the evolving landscape of RAG system research.

Author: Sigita Lapiņa

Supervisor: Dmitry Pavlyuk

Degree: Master

Year: 2024

Work Language: English

Study programme: Computer Sciences

More...

Development of Cross Platform 2D Games using OpenGL and ImGui Libraries

The aim of this bachelor's work was to develop a cross platform application containing two 2D games - Tetris and Snake. In the course of the work the subject area was analyzed and present analogues compared. The requirements were formulated based on the analysis conclusions. Soon after the model of requirements was created where the actors were identified and use cases developed. Based on the developed use cases a Use Case diagram was built as well as sequence diagrams and preliminary class diagram. During implementation stage, the software was developed in C++ language using OpenGL and Dear ImGui graphical libraries. In the end 2 games were developed. Finally, the developed software was tested. Test results showed that developed application in the process of completing the bachelor's work fully complies with all the requirements for it.

Author: Staņislavs Ņilovs

Supervisor: Karina Kostjkina

Degree: Bachelor

Year: 2024

Work Language: English

Study programme: Computer Science

More...

CNN-Based pipeline-related artifact and damage recognition in IHC staining as preprocessing step for pathological analysis

This work proposes automated solution for artifact and damage segmentation in biomedical images using machine learning algorithms. The development process includes data preprocessing, label classification using a clustering algorithm and segmentation model. CNN architectures like YOLO and U-NET are utilized for segmentation, and K-Means and DBSC algorithms are evaluated for clustering. The outcomes include a set of data preprocessing precodures, clustering algorithm testing and results analysis, segmentation model and recommendations for further development.

Author: Taisija Kožarina

Supervisor: Jeļena Kijonoka

Degree: Bachelor

Year: 2024

Work Language: English

Study programme: Computer Science

More...

Research on Software Development Aspects Using node.js Technology

To explore the aspects of software development, an application was created using Node.js and NestJS to build a REST API. This API integrates Google Natural Language AI to analyze submitted user reviews. The application includes multiple REST API endpoints that can process, analyze, and aggregate user review data. Performance measurements were conducted, analyzing event loop latency, memory and CPU usage, and other key metrics. MongoDB was used for data storage. The work also includes API and its performance evaluation. The system was tested to ensure it meets the set criteria and provides practical application.

Author: Valērijs Sergejevs

Supervisor: Mihails Savrasovs

Degree: Bachelor

Year: 2024

Work Language: English

Study programme: Computer Science

More...

Table View
Text View