Institutional Repository of Vilnius University
Automobilių registracijos numerių atpažinimas /
Abstract
dc:descriptionThe ever-increasing desire of people to automate as many processes as possible is inevitably linked to automatic character recognition as well. More and more entrepreneurs or various institutions are using equipment that allows them to automatically and efficiently scan car number plates, convert them into plain text and continue to use them for a variety of purposes. Optical character recognition programs, which consist of several key processes, such as image pre-processing and character recognition algorithms, are ideal for this purpose. However, sometimes converting characters from a photo to plain text can be quite challenging. Noise in the photo, such as sand, dust, rain or twilight can affect the quality of the photo and the accuracy of character recognition. This paper analyzes the methods of primary photo processing, investigates which of them work optimally and helps to improve the accuracy of character recognition of the car registration plate when working with photos taken with a smartphone and in good lighting. Also, the paper discusses the character recognition tools Tesseract and Amazon Rekognition - the principles of their work and the use of neural networks are reviewed. Experimentally, an attempt is made to determine the optimal methods and their parameters of photos pre-processing, as well as, there is little mention of debugging the character recognition tool calling function. After conducting the study and adapting the primary photo processing functions to the primary data set (140 photos), an experiment was performed with 20 photos of random car registration numbers. It turned out that it is optimal to apply several variations of pre-processing algorithms with different image resizing, noise reduction, morphological and other functions and their parameters for character recognition. Using combined algorithms in the program, 90 percent accuracy was achieved for the data taught and 95 percent accuracy for the random data.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Meška, Jonas,
- Contributors dc:contributor
-
- Kybartas, Rimantas
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD81805718&prefLang=en_US