We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Text localization in digital images using a hybrid method.
- Authors
Akoushideh, Alireza; Rasoulnejad, Sayed Mohammad Fallah; Shahbahrami, Asadollah
- Abstract
Text localization in digital images includes various applications such as routing in auto-driving, postal services, identifying containers in ports, robotic navigation in urban environments, and so on. Multiple methods such as Stroke Width Transform (SWT), Local Adaptive Thresholding (LAT), and Maximally Stable Extremal Regions (MSER) have already been noticed by many researchers. These methods address specific challenges such as text orientation, font variations, size, and non-uniform illuminations. Additionally, deep methods with high accuracy are not optimal regarding computation time and data required for training. Our research considers the SWT, LAT, and MSER methods as a pre-processing step with some improvement techniques to simultaneously deal with the challenges above and improve the text localization result. Maximally Stable Extremal Regions, multilevel binarization, and non-maximum suppression techniques have been used in our proposed approach. Our proposed approach has reasonable results compared with the none-deep state-of-the-art algorithms and deep learning-based methods on the ICDAR 2013 dataset. Recall, precision, and f-measure are 0.7442, 0.9116, and 0.8195, respectively.
- Subjects
DIGITAL images; CONTAINER terminals; TEXT messages; POSTAL service
- Publication
Multimedia Tools & Applications, 2022, Vol 81, Issue 23, p34047
- ISSN
1380-7501
- Publication type
Article
- DOI
10.1007/s11042-022-13179-2