Tamil character recognition from ancient epigraphical inscription using OCR and NLP

Recognition of ancient Tamil characters is one of the challenging task for Epigraphers as the language has evolved with different characters set. If the inscriptions are on stone walls, it adds even more complexity in identifying characters. This proposed work mainly focuses on recognition of various Tamil characters between 9th and 12th centuries using OCR and NLP techniques. In this work, the inscription images collected from Tamil Nadu, Archaeological Department are pre-processed and segmented. During the segmentation process the color images were converted to gray image and to binary image based on threshold value. From segmented, image features like number of lines, curves, loops and dots have been extracted using Scale Invariant Feature Transform (SIFT) algorithms for each letter to identify the exact character. Characters will be classified and constructed based on Vectors extracted, using Support Vector Machine (SVM) classifier and the patterns of the character will matched with known characters and predicted using Trigram technique. Each identified character will be assigned with its corresponding Unicode value and it will be updated in the image corpus for further character identification, and to make the system in identifying the characters more effectively. Thus the proposed system can solve the major problems in reading the inscription images.