DETECTION OF TEXT OBJECTS IN IMAGES OF REAL SCENES BASED ON CONVOLUTIONAL NEURAL NETWORK MODEL
Abstract
A model of text image detector based on a convolutional neural network architecture is presented, capable of synthesizing high-level features of images in the «black box» mode. An implementation of the detector application, based on algorithms of multi-scale scanning and local responses interpretation is described, allowing to find out text samples on images of real scenes. Advantages in comparison with analogs are shown and efficiency evaluation on an example of a known database is conducted.
References
1. Sumathi, C.P. A Survey on various approaches of text extraction in images / C.P. Sumathi,
2. T. Santhanam, G. Gayathri // International Journal of Computer Science & Engineering Survey. – 2012. – Vol. 3, № 4. – P. 27–42.
3. LeCun, Y. Gradient-Based Learning Applied to Document Recognition / Y. LeCun, L. Bottou // Proceedings of the IEEE. – 1998. – Vol. 86, № 11. – P. 2278–2324.
4. Кузьмицкий, Н.Н. Сверточная нейросетевая модель в задаче классификации изображений изолированных цифр / Н.Н. Кузьмицкий // Доклады БГУИР. – Минск, 2012. – № 7. – С. 64–70.
5. Головко, В.А. Нейронные сети: обучение, организация и применение : учеб. пособие / В.А. Головко. – М. : ИПРЖР, 2001. – Кн. 4. – 256 с.
6. Осовский, С. Нейронные сети для обработки информации / С. Осовский. – М. : Финансы и статистика, 2002. – 344 с.
7. Delakis, М. Text detection with convolutional neural networks / М. Delakis, Сr. Garcia // Intern. Conf. on Computer Vision Theory and Applications. – Cambridge, 2008. – P. 290–294.
8. Wang, K. End-to-end scene text recognition / K. Wang, B. Babenko, S. Belongie // IEEE Intern. Conf. on Computer Vision (ICCV). – Barcelona, 2011. – P. 1457–1464.
9. ICDAR 2003 robust reading competitions / S.M. Lucas [et al.] // Proc. of Seventh Intern. Conf. on Document Analysis and Recognition. – Edinburgh, 2003. – P. 682–687.
10. Campos, T.E. Character Recognition in Natural Images / T.E. Campos, B.R. Babu // VISAPP. – 2009. – Vol. 2. – P. 273–280.
11. Touch TT : Scene text extractor using touchscreen interface / J. Jung [et al.] // ETRI Journal. – 2011. – Vol. 33, № 1. – P. 78–88.
12. The Street View House Numbers (SVHN) Dataset [Electronic resource]. – 2011. – Mode of access : http://ufldl.stanford.edu/housenumbers. – Date of access : 03.07.2014.
13. Ikica, A. An improved edge profile based method for text detection in images of natural scenes / A. Ikica, P. Peer // Intern. Conf. on Computer as a Tool (EUROCON). – Lisbon, 2011. – P. 1–4.
14. ICDAR 2013 Robust Reading Competitio / D. Karatzas [et al.] // Proc. 12th Intern. Conf. of Document Analysis and Recognition, IEEE CPS. – Washington, 2013. – P. 1115–1124.
15. Wolf, C. Object Count Area Graphs for the Evaluation of Object Detection and Segmentation Algorithms / C. Wolf, J.M. Jolion // International Journal of Document Analysis. – 2006. – Vol. 8, № 4. – P. 280–296.
Review
For citations:
Kuzmitsky N.N. DETECTION OF TEXT OBJECTS IN IMAGES OF REAL SCENES BASED ON CONVOLUTIONAL NEURAL NETWORK MODEL. Informatics. 2015;(2):12-21. (In Russ.)
Views: 972
Email this article 