350 rub
Journal Electromagnetic Waves and Electronic Systems №3 for 2017 г.
Article in number:
Cascade classifiers application for solving character classification problem
Authors:
E.A. Bobyleva - Post-graduate Student, «Computer systems and networks», Kaluga branch of the Bauman MSTU E-mail: foxymoon@yandex.ru V.E. Drach - Ph. D. (Eng.), Associate Professor, Department «Design and manufacturing of electronic equipment», Kaluga branch of the Bauman MSTU E-mail: drach@bmstu-kaluga.ru A.V. Rodionov - Ph. D. (Eng.), Associate Professor, Department «Computer systems and networks», Kaluga branch of the Bauman MSTU E-mail: andviro@gmail.com I.V. Chukhraev - Ph. D. (Eng.), Associate Professor, Head of Department «Computer systems and networks», Kaluga branch of the Bauman MSTU E-mail: igor.chukhraev@mail.ru
Abstract:
Text and character detection and classification in natural scene images is one of the perspective research tendencies. This article describes cascade LBP-classifiers application for detection and classification of the Japanese characters of syllabic alphabets in both the natural scene images and the text documents. In this research, cascade classifiers were trained using different sets of positive and negative images. The set of positive images consisted of printed and calligraphic characters. Experiments were performed in two stages. In the first stage, three sets of negative images were used. First set consisted of natural scene images, second set consisted of characters belonged to two other Japanese alphabets, third consisted of natural scene images and Japanese alphabet characters. In the second stage, sets of 10000 positive and 5000 negative samples and 20000 positive and 10000 negative samples were used. Experiments showed that different sets of negative images may affect on false positive detection and classification accuracy, but the increase of training samples number leads to the rise of classifier invariance and the decrease of false positive detection amount. Thus, the detection and false positive rates of the set of negative images consisted of natural scene images and Japanese alphabet characters and set of 20000 and 10000 positive and negative samples are 75−77% and 4−5% respectively.
Pages: 61-66
References

 

  1. Viola P., Jones M. Rapid object detection using a boosted cascade of simple features // Computer Society Conference on Computer Vision and Pattern Recognition. 2001. V. 1. P. 511−518.
  2. Liao S., Zhu X., Lei Z., Zhang L., Li S. Learning multi-scale block local binary patterns for face recognition // Advances in Biometrics. 2007. V. 4642. P. 828−837.
  3. Churchill M., Fedor A. Histogram of Oriented Gradients for Detection of Multiple Scene Properties. 2015.  URL = http://web.stanford.edu/class/cs231a/prev_projects/CS231AHOGFinalReport.pdf (data obrashhenija: 17.02.2017).
  4. Chen X, Yuille A. Detecting and reading text in natural scenes // Computer Vision and Pattern Recognition. 2004. V. 2. P. 366−373.
  5. Jiang R., Qi F., Xu L., Wu G. Detecting and Segmenting Text from Natural Scenes with 2‑Stage Classification // Intelligent Systems Design and Applications. 2006. V. 1. P. 819−824.
  6. Escalera S., Baró X., Vitrià J., Radeva P. Text Detection in Urban Scenes // Artificial Intelligence Research and Development. 2009. P. 35−44.
  7. Opitz M. Text Detection and Recognition in Natural Scene Images: master thesis // Computer Vision Lab Institute of Computer Aided Automation Vienna University of Technology. 2013.
  8. Tian S., Pan Y., Huang C., Lu S., Yu K., Tan C.L. Text Flow: A Unified Text Detection System in Natural Scene Images // International Conference on Computer Vision. 2015. P. 4651−4659.
  9. Zhu S., Zanibbi R. A Text Detection System for Natural Scenes With Convolutional Feature Learning and Cascaded Classification // Computer Vision and Pattern Recognition. 2016. P. 625−632.
  10. Yin X.C., Yin X., Huang K., Hao H.W. Robust Text Detection in Natural Scene Images // Transactions on Pattern Analysis and Machine Intelligence. 2014. V. 36. № 5. P. 970−983.
  11. Das S., Banerjee S. An Algorithm for Japanese Character Recognition // International Journal of Image, Graphics and Signal Processing. 2015. V. 7. № 1. P. 9−15.
  12. Tsai C. Recognizing Handwritten Japanese Characters Using Deep Convolutional Neural Networks. 2016. URL = http://cs231n.stanford.edu/reports2016/262_Report.pdf (data obrashhenija: 17.02.2017).
  13. Rødland T. Classifying Glyphs. Comparing Evolution and Learning: master thesis // Norwegian University of Science and Technology. 2011.
  14. JAsinskijj F.N., Mochalov A.S. Raspoznavanie bolshogo chisla obrazov pri pomoshhi nejjronnykh setejj s ispolzovaniem mnogoprocessornykh sistem // Vestnik Ivanovskogo gosudarstvennogo ehnergeticheskogo universiteta. 2011. № 2. S. 85−87.
  15. Belevskijj V.A., Maksimov A.V. Metody invariantnogo raspoznavanija kontrastnykh cherno-belykh izobrazhenijj simvolov russkogo alfavita // Voprosy radioehlektroniki. Ser. EHVT. 2010. T. 3. № 4. S. 5−24.
  16. Awazu T., Fukuo M., Takata M., Joe K. A Multi-fonts Kanji Character Recognition Method for Early-modern Japanese Printed Books with Ruby Characters // Pattern Recognition Applications and Methods. 2014. P. 637−645.
  17. Fukuo M., Enomoto Y., Yoshii N., Takata M., Kimesawa T., Joe K. Evaluation of the Svm Based Multi-fonts Kanji Character Recognition Method for Early-modern Japanese Printed Books // Parallel and Distributed Processing Techniques and Applications. 2011. V. 2. P. 727−732.
  18. Guennouni S., Ahaitouf A., Mansouri A. A Comparative Study of Multiple Object Detection Using Haar-Like Feature Selection and Local Binary Patterns in Several Platforms // Modelling and Simulation in Engineering. 2015. V. 2015.
  19. Bulatnikov E.V., Goeva A.A. Sravnenie bibliotek kompjuternogo zrenija dlja primenenija v prilozhenii, ispolzujushhem tekhnologiju raspoznavanija ploskikh izobrazhenijj // Vestnik MGUP imeni Ivana Fedorova. 2015. № 6. S. 85−91.