Script identification from handwritten documents using SIFT method

Automatic identification of scripts from document images helps selecting appropriate OCR for character recognition and content retrieval. In this paper, Scale invariant Feature Transformation (SIFT) based script identification has been proposed. Features are extracted using SIFT approach at word level (two, three or more character words) and KNN classifier has been used to recognize the script. Experiments are performed by extracting the words from document images consisting of English, Kannada, and Devanagari scripts. Overall accuracy reported for the proposed system is 97.65% and 96.71% for bi-script and tri-scripts, respectively.