Article Preview
Top1. Introduction
OCR systems are script specific in a sense that they are made to read characters written in a particular script only. Script is defined as the graphic form of any writing system which is used to express the written languages (Singh et al., 2015). Languages throughout the world are typeset in many different scripts. For example, Devanagari is used for writing a number of Indian languages like Hindi, Konkani, Sanskrit, Nepali, etc., whereas Assamese and Bengali languages use different variants of the Bangla script. The documents like language translator, certain official application forms, examination question papers, etc. contain words written in more than one script. Therefore, in this multilingual and multi-script world, OCR systems need to be capable of recognizing characters irrespective of the script in which the documents are written. If all the characters of different scripts are handled simultaneously at the classification stage, then the accuracy of the overall OCR system would surely decrease. So, for successful implementation of OCR system for multi-script documents, it is necessary to assemble different documents script-wise before feeding to the individual OCR system. This gives rise to the problem of script identification which is usually defined as the process of identifying scripts in any multi-script environment so that the recognized scripts can be sent to their corresponding OCR engines for recognition purpose.
Automatic script identification facilitates sorting, searching, indexing, and retrieving of multilingual documents. Script recognition also helps in video indexing and retrieval when dealing with a multi-script content. So, the manifold applicability along with its challenges of the script identification problem inspires the researcher a lot. However, research in script identification is still in its early stage.
The problem of script identification is generally achieved at any of the three levels: (a) Page-level, (b) Text line-level, and (c) Word-level. Performing script recognition at word-level is much more challenging task than at the other two higher levels. The primary reason for this is the information congregated from the characters constitutes the word image, sometimes may not be enough to identify the script truthfully. In addition to this, the presence of noise, multi-level skewness or slant nature, non-uniform gaps between inter-words (or intra-words) and other significant degradations also significantly affect script identification procedure at word-level. Again, the recognition of scripts at word-level also requires automation of grouping words belonging to a specific script/language and also separate interlaced words pertaining to other scripts. In contrast, the computational complexity of feature extraction at page-level and text line-level can be sometimes too laborious and time-consuming. Keeping the facts in consideration, it is decided that script identification at word-level is a fair option among three alternatives.
The rest of paper is organized as follows: The state-of-the-art methods related to script identification are detailed in Section 2 whereas some basic information related to the scripts used in the present work is described in Section 3. The proposed method is presented in Section 4. Experimental results and their analyses are given in Section 5. Section 6 concludes the work and lists some future directions.