Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font Diversity
Reading Indian scene texts is complex due to the use of regional vocabulary, multiple fonts/scripts, and text size. This work investigates the significant differences in Indian and Latin Scene Text Recognition (STR) systems. Recent STR works rely on synthetic generators that involve diverse fonts to...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-03-01
|
| Series: | Journal of Imaging |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2313-433X/8/4/86 |
| _version_ | 1827619638593191936 |
|---|---|
| author | Sanjana Gunna Rohit Saluja Cheerakkuzhi Veluthemana Jawahar |
| author_facet | Sanjana Gunna Rohit Saluja Cheerakkuzhi Veluthemana Jawahar |
| author_sort | Sanjana Gunna |
| collection | DOAJ |
| description | Reading Indian scene texts is complex due to the use of regional vocabulary, multiple fonts/scripts, and text size. This work investigates the significant differences in Indian and Latin Scene Text Recognition (STR) systems. Recent STR works rely on synthetic generators that involve diverse fonts to ensure robust reading solutions. We present utilizing additional non-Unicode fonts with generally employed Unicode fonts to cover font diversity in such synthesizers for Indian languages. We also perform experiments on transfer learning among six different Indian languages. Our transfer learning experiments on synthetic images with common backgrounds provide an exciting insight that Indian scripts can benefit from each other than from the extensive English datasets. Our evaluations for the real settings help us achieve significant improvements over previous methods on four Indian languages from standard datasets like IIIT-ILST, MLT-17, and the new dataset (we release) containing 440 scene images with 500 Gujarati and 2535 Tamil words. Further enriching the synthetic dataset with non-Unicode fonts and multiple augmentations helps us achieve a remarkable Word Recognition Rate gain of over <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>33</mn><mo>%</mo></mrow></semantics></math></inline-formula> on the IIIT-ILST Hindi dataset. We also present the results of lexicon-based transcription approaches for all six languages. |
| first_indexed | 2024-03-09T10:32:54Z |
| format | Article |
| id | doaj.art-45989a9a86664e9988278ad7dc0d8e07 |
| institution | Directory Open Access Journal |
| issn | 2313-433X |
| language | English |
| last_indexed | 2024-03-09T10:32:54Z |
| publishDate | 2022-03-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Journal of Imaging |
| spelling | doaj.art-45989a9a86664e9988278ad7dc0d8e072023-12-01T21:07:32ZengMDPI AGJournal of Imaging2313-433X2022-03-01848610.3390/jimaging8040086Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font DiversitySanjana Gunna0Rohit Saluja1Cheerakkuzhi Veluthemana Jawahar2Centre for Vision Information Technology, International Institute of Information Technology, Hyderabad 500032, IndiaCentre for Vision Information Technology, International Institute of Information Technology, Hyderabad 500032, IndiaCentre for Vision Information Technology, International Institute of Information Technology, Hyderabad 500032, IndiaReading Indian scene texts is complex due to the use of regional vocabulary, multiple fonts/scripts, and text size. This work investigates the significant differences in Indian and Latin Scene Text Recognition (STR) systems. Recent STR works rely on synthetic generators that involve diverse fonts to ensure robust reading solutions. We present utilizing additional non-Unicode fonts with generally employed Unicode fonts to cover font diversity in such synthesizers for Indian languages. We also perform experiments on transfer learning among six different Indian languages. Our transfer learning experiments on synthetic images with common backgrounds provide an exciting insight that Indian scripts can benefit from each other than from the extensive English datasets. Our evaluations for the real settings help us achieve significant improvements over previous methods on four Indian languages from standard datasets like IIIT-ILST, MLT-17, and the new dataset (we release) containing 440 scene images with 500 Gujarati and 2535 Tamil words. Further enriching the synthetic dataset with non-Unicode fonts and multiple augmentations helps us achieve a remarkable Word Recognition Rate gain of over <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>33</mn><mo>%</mo></mrow></semantics></math></inline-formula> on the IIIT-ILST Hindi dataset. We also present the results of lexicon-based transcription approaches for all six languages.https://www.mdpi.com/2313-433X/8/4/86scene text recognitiontransfer learningphoto OCRmulti-lingual OCRIndian languagesindic OCR |
| spellingShingle | Sanjana Gunna Rohit Saluja Cheerakkuzhi Veluthemana Jawahar Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font Diversity Journal of Imaging scene text recognition transfer learning photo OCR multi-lingual OCR Indian languages indic OCR |
| title | Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font Diversity |
| title_full | Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font Diversity |
| title_fullStr | Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font Diversity |
| title_full_unstemmed | Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font Diversity |
| title_short | Improving Scene Text Recognition for Indian Languages with Transfer Learning and Font Diversity |
| title_sort | improving scene text recognition for indian languages with transfer learning and font diversity |
| topic | scene text recognition transfer learning photo OCR multi-lingual OCR Indian languages indic OCR |
| url | https://www.mdpi.com/2313-433X/8/4/86 |
| work_keys_str_mv | AT sanjanagunna improvingscenetextrecognitionforindianlanguageswithtransferlearningandfontdiversity AT rohitsaluja improvingscenetextrecognitionforindianlanguageswithtransferlearningandfontdiversity AT cheerakkuzhiveluthemanajawahar improvingscenetextrecognitionforindianlanguageswithtransferlearningandfontdiversity |