From Stone Inscriptions to Digital Text: Enhancing Archaeological Scripts and Automated Conversion

Abstract

Analysis of archeological inscription plays a crucial role in understanding the values of historical, cultural and heritage information. Many of the available inscriptions are damaged and deteriorated in color and clarity, thus the identifications become a challenge. In this work, Various filtering techniques used to enhance the Tamil stone inscription images are analyzed empirically. Performances of filtering techniques are compared and evaluated in terms of certain features of an image, specifically the Peak Signal-to-Noise Ratio (PSNR)value, Signal-to-Noise Ratio (SNR) and Mean Squared Error (MSE) value. The effectiveness and efficiency of these techniques are tested using different categories of input images; thereby the most efficient filtering technique can be determined based on the image. As the ancient inscriptions are of different formats, we have devised a conversion system which converts Brahmi scripts to modern Tamil script. OCR is used to recognize the Brahmi characters and converted into Tamil characters using code point toning technique. The proposed system uses Tesseract OCR, which has been trained manually for the recognition of Tamil Brahmi Letters. The core functionality of the OCR depends on LSTM. LSTM cells store information about the input for a specified time period which makes it suitable for language training. Further to transcript the recognized 1 Brahmi letters to present age Tamil letters, code points are toned with the present age Tamil letters using Unicode-Character lookup table. The overall character and word error rate of the system has been reduced to 0.0086 and 0.057%.