A Copula Based Statistical Model for Text Extraction from Scene Images
This article proposes a scheme for automatic text extraction from scene images. The work is composed of two steps. In the first step, we apply a model based color segmentation procedure in the LCH color space. This step produces certain homogenous connected components (CCs) from the image. In the next step, these CCs are examined in order to identify possible text components. A number of features that distinguish between text and non-text components, are defined. Further, during learning, these features are considered independently and approximated using parametric distribution families. Finally, the joint distribution of the features are constructed using a multivariate Gaussian copula. Consequently, we obtain two copula based class distributions for the two classes (text and non-text). Afterwards, during testing, a CC belongs to the class that produces the highest class distribution probability. Our experiments are on the database of ICDAR 2003 Robust Reading Competition. The experimental results are satisfactory.
KeywordsText Component Tail Dependence Scene Image Gaussian Copula Color Image Segmentation
- 2.Jung, K., Kim, I.K., Kurata, T., Kourogi, M., Han, H.J.: Text scanner with text detection technology on image sequences. In: Proc. of Int. Conf. on Pattern Recognition, vol. 3, pp. 473–476 (2002)Google Scholar
- 3.Ghoshal, R., Roy, A., Parui, S.K.: Recognition of bangla text from scene images through perspective correction. In: Proc. ICIIP, pp. 385–390 (2011)Google Scholar
- 4.Mardia, K.V., Jupp, P.: Directional Statistics. John Wiley and Sons Ltd. (2000)Google Scholar
- 7.Roy, A., Parui, S.K., Paul, A., Roy, U.: A color based image segmentation and its application to text segmentation. In: Proc. of Ind. Conf. on Computer Vision, Graphics & Image Processing, pp. 313–319 (2008)Google Scholar
- 8.Nelsen, R.B.: An Introduction to Copulas. Springer (2006)Google Scholar