Simon
11 lines
2.0 KiB
TeX
11 lines
2.0 KiB
TeX
\chapter{Abstract}
|
|
\begin{english}
|
|
|
|
This bachelor's thesis focuses on one of the many fields of optical character recognition: the extraction of textual data from digital screenshots. The goal is to maximize the amount and quality of the resulting data, hence simplifying the management of graphical resources for COPA-DATA's product documentation.
|
|
|
|
For this purpose, a selection of algorithms including resampling or binarization for image processing is chosen. For filtering the resulting data, natural language processing techniques, for instance normalization or basic language-specific filtering methods are being selected. After explaining the algorithms in their basic function, they are objectively compared using common metrics for speech and text recognition.
|
|
|
|
Analyzing different pictures using just one method does not always lead to satisfying results. To avoid losing important details, specifically when working with binarization techniques such as thresholding, the chosen parameters must match the individual characteristics of the picture. Based on the comparisons conducted in this bachelor thesis, the text recognition system for the COPA-DATA product documentation can be parameterized in such a way that it captures as many details as possible when processing different input images, optimizing the end result of text recognition within the image.
|
|
|
|
The prototypical implementation and its respective components are designed to be reusable for further research or adaptation to specific requirements. Due to the modular structure of the automated comparison system, it is possible to add new functionality and to edit existing features. As a result, a satisfying text recognition approach can always be determined with little effort, even after changing the display language or after a complete redesign of the graphical user interface. Further steps to improve text recognition, such as edge detection in image processing or "fuzzy matching" to capture result data, are feasible due to the flexible structure.
|
|
\end{english} |