A system for word retrieval from Kannada documents

In this paper, we propose a technique for retrieval of printed Kannada words from a digital repository based on Gabor wavelets and structural features. Gabor wavelets are employed to capture global properties of the underlying image whereas structural features are used to extract the local properties. We call the combination of these features as Glocal. An input document image is segmented into words and stored into a library. Then, Glocal features are employed to represent the words. Next cosine distance is used to measure the similarity between two words, based on it; relevance of the word is estimated by generating distance ranks. Then correctly matched words are selected at different distance thresholds such as 96%, 97%, 98% and 99%. Encouraging results are achieved in terms of average precision rate 74.27%, average recall rate 79.26% and F measure 75.62 % at a threshold of 97%.