Enhanced Preprocessing Technique for Degraded Printed Marathi Characters

Abstract

Designing a dynamic and proficient OCR entity is an interesting and captivating region in image processing. The OCR organism purposes to decipher text in images to a machine comprehensible text. A successful degraded document has numerous applications like preservation of history, persistent documents, and many more. For the same, copious policies occur for numerous scripts, languages, and so far for virtuous class papers. Contrariwise, unique restricted versions have been explored for degraded printed Marathi characters. The OCR structure comprises stages like data collection, preprocessing, feature extraction, segmentation, classification, and recognition. For this work, we have developed our dataset size of 4900 images for 49 isolated Marathi characters. Among these, preprocessing phase plays a significant role, especially in the case of degraded characters. This paper focuses work on preprocessing of degraded Marathi characters. Mean Square Error, Mutual information, and Peak signal to noise ratio assessment factors are used for the evaluation of the enriched image. The proposed approach’s effects are attained in MATLAB R2015a. Many researchers have made efforts for developing OCR for degraded printed documents for various languages, whereas very less attempts made for degraded Marathi character recognition. In the case of the degraded Marathi character recognition problem, the recognition accuracy is largely dependent on the preprocessing techniques and how the character information and features are retained. To overcome the limitations of earlier reported systems of loss of information and features retention, a novel and enhanced preprocessing technique is developed specifically for degraded Marathi character recognition. The proposed methodology is giving promising results as compared to other methods which are 35.14. Similarly, the average mutual information value of the proposed technique is 2.71 which is greater than others. The projected practices offered less value of average mean square error and response time. The proposed method for preprocessing will improve the recognition accuracy for degraded isolated Marathi characters.