Abstract: Two novel techniques for language identification of both, machine printed and handwritten document images, are presented. Language identification is the procedure where the language of a given document image is recognized and the appropriate language label is returned. In the proposed approaches, the main body size of the characters for each document image is determined, and accordingly, a sliding window is used, in order to extract the SIFT local features. Once a large number of features have been extracted from the training set, a visual vocabulary is created, by clustering the feature space. Data clustering is performed using K-means or Gaussian Mixture Models and the Expectation - Maximization algorithm. For each document image, a Bag of Visual Words or Fisher Vector representation is constructed, using the visual vocabulary and the extracted features of the document image. Finally, a multi class Support Vector Machine classification scheme is used, to score the system. Experiments are performed on well-known databases and comparative results with another established technique, are also given.

Nikos Tatarakis and Ergina Kavallieratou. “Language Identification by Using SIFT Features”. International Journal of Advanced Research in Artificial Intelligence (IJARAI) 4.12 (2015). http://dx.doi.org/10.14569/IJARAI.2015.041206

@article{Tatarakis2015,
title = {Language Identification by Using SIFT Features},
journal = {International Journal of Advanced Research in Artificial Intelligence},
doi = {10.14569/IJARAI.2015.041206},
url = {http://dx.doi.org/10.14569/IJARAI.2015.041206},
year = {2015},
publisher = {The Science and Information Organization},
volume = {4},
number = {12},
author = {Nikos Tatarakis and Ergina Kavallieratou}
}