For verification of this paper, please visit on http://www.socialresearchfoundation.com/innovation.php#8

Marathi is Indo-Aryan language spoken about 83 million people in India’s central and western region. In 1966, it was designated the official language of the state of Maharashtraof India’s 22 scheduled languages. It’s geographical range stretches from the north western part of Mumbai apart the western coast through goa and eastward over the deccan. There were approximately 83 million native Marathi language speakers as per 2011. Many Inscription have been found in ancient Indian cities[1], These inscriptions are very important and important documents of ancient India to learnhistorical information on the time and area, cases of the evolution of language over the centuries[2].

Review of Marathi Language:

There are types of marathi among all the languages in the world, ranking tenth in the world. Marathi also has some ancient literature. Marathi is one of the language of Sanskrit origin. The marathi language first appears in the stone inscription originate from the 11^th century in from the 13^thto 20^th century nearing it’s middle, year were written in the modi script, and from 1950 to Sanskrit . use language as a communication tool. Article while the purpose of language is to express and understand the thoughts of other, writing is to write characters from two or more language.Marathi is written is form of Devanagari script, Marathi consist of 12 vowels, 36 consonants and numbers are given in figure1,2 and 3 respectively.

A programme that can read characters from old Sinhala inscriptions using optical character recognition(OCR). The OCR module makes use of artificial neural network(ANN) and convolutional neural network(CNN) technology. The recognition rates using genius test photos, pre-processed test data, and from training are all included in the evaluation. The selected OCR solution in CNN since it performs better than ANN. CNN can only efficiently identify a character since there is not enough data[3].The article reviews methods for deciphering ancient marathi scripts from stone inscription, focuses on difficulties like background noise and script background similarly doesn’t provide any concerts finding but makes the case for future research on enhancing the precision of recognition[4].Abhishek Tomer, Minu Choudhary, Amit Yerpude[5] by comprehending the script and language in the photo, the author performed a survey investigation central on Indian inscription identification of characters areas and carried out prehistoric inscription analysis.Anush Goel and Akash Sehrawat [6] proposed a system that provides automatic data reading for blind people using Raspberry Pi. uses OCR technology to recognize printed characters using image detection devices and python programming.

Convert images to audio format using OCR and Text-to-Speech (TTS). The diagram for the conversion was made using a Raspberry Pi and the was reworked using the Tesseract library and python. Section Use the OpenCV library to process the text and complete the audio output.Shuping Liu, Yantian Xian, Huafeng Li, and Zentao Yu report a new detection model based on MCA and different representations. It provides two useful results for text search. The first is to develop the capacity to summarize and search for articles using the MCA method in summarizing. Article The method used for document search improves document search performance [7].Ancientdocumentsrecognition using statistical feature extraction techniques Preprocessing (Noise Removal, Binarization, etc) using local or global Thresholding Feature extraction CNN(88.95), section of the characters frame, including overlapping and non-overlapping zones, are divided up. Cone pixels may be used to calculate density, which is calculate as the whole number of pixels divided by the number of pixels in each zone[8].it is the next level of segmentation.

In their 2015 research, Sachin S. Bhat and H.V. Balachandra Achar sought to use cutting-edge identification algorithms to pinpoint the historical era of various ancient kannada scripts. Their technique included elements for image processing such character set segmentation, noise reduction, feature extraction, classification, and characters from inscriptions was an inventive element. Their MATLAB-based studies, which had an accuracy rate of 80%, proved the viability of their strategy[9].Soumya A and G Hemantha Kumar research focuses on improving the creation of ancient manuscripts using preprocessing and segmentation techniques. They convert scanned images into images in digital format. To enhance results, they present three filtering techniques: Gaussian blur, Laplacian filter, and USM filter. To get better result in terms of image quality and classification accuracy, these filters can be used in variety of combinations, including non-uniform filters[10].

Inscription :

The information history or information written on brickwork, stone and other brittle surfaces is call inscription. Indian Inscription may be categorised into four groups : stone, cave, pillar and wall. The written worlds on cave walls are called cave inscription. information written on originally stones is known as stone inscription. later inscription or pillar inscription were curved on artificially finished stone slabs, pillar and other stone materials.The figure 4 is one of the oldest known marathi inscription was unearthed at the foot of the Bahubali statue at the jain temple at Shravanabelagola.

Fig 4. Marathi inscription

Architecture  :

This section provides a succinct overview of the general structure of the architecture of converting marathiinscript to marathi text, as shown in figure 5. Which converts marathi in script to marathi text, which also stores Marathi text as output is stored in a text file(given a text file name).The architecture divided mainly into three part namely a. Pre-processing module, b.Text recognition module, and c. Post-processing module

Fig 5. Architecture of converting Marathiinscript to Marathi text

1. Pre-processing module : the pre-processing module is responsible for preparing the input image before it is fed into the text module. it’s main purpose is to improve image quality and improve OCR accuracy. Here are some preliminary steps are Image Resizing : Increase image size for better performance, Binary : simplify text extraction by converting images to binary format,Noise Reduction : Removes unused noise and imperfection in image that can be caused by imaging.

2. Text recognition module :The module for text recognition is the core of the OCR system. It will use the pre-processed image as input and perform correct character/word recognition. There are many ways to recognize text, including, Optical Character Recognition(OCR) : OCR algorithms analyze images to recognize & identify characters in text, Documents layout Analysis: understanding the layout of documents to determine the reading and distribution of documents.Modern text recognition after uses deep learning techniques such as convolutional neural network (CNNs) to achieve character recognition accuracy.

3. Post-processing module : the post-processing module takes the output recognition by the text module and further optimizes it to improve overall accuracy & fix anything not visible in the test at confirmation. Finishing operation include: Language model community, Context analysis, Confidence score.The post-processing module is required to solve OCR problem caused by various factors such as poor image quality, hard text, or invisible characters.

The user interface is developed by using Python as front end and OCR as backend where it enables computer to perform activities of reading Marathi inscription and convert it to equivalent Marathi Text as shown in figure 6, 7 and 8 respectively: Text detection process consist of Pre-processing:  The image concept is pre-processed to improve its quality and eliminate noise, which will improve the work quality of the next step,edge Detection: Use an edge detection technique to detect high contrast areas often associated with text borders. Text Region Suggestion: Use different methods such as floating windows or anchors to show regions containing text. Text Field Validation: These defined fields will be subjected to a validation process to determine if they contain text. This step usually involves classifying text and nontextual regions using machine learning or deep learning models.Text Localization: After verifying a region of text, a bounding box or pattern is drawn around the text to determine its position.

1. K.G.N.D. Karunarathne1, K.V. Liyanage2, D.A.S. Ruwanmini3, G.K.A. Dias4, S.T. Nandasara5," Recognizing ancient Sinhala Inscription Characters using Neural Network Technologies” International Journal of Scientific Engineering and Applied Science (IJSEAS) –ISSN: 2395-3470 3(1), Jan 2017.(https://ijseas.com/volume3/v3i1/ijseas20170104.pdf)

2. P.Nikhi,V.Jayakumar, S.kolkure [2015] OPTICAL CHARACTER RECOGNITION: AN ENCOMPASSING REVIEW [ONLINE] Available at: http://esatjournals.net/ijret/2015v04/i01/IJRET20150 401062.pdf

3. K.G.N.D. Karunarathne, K.V. Liyanage, D.A.S. Ruwanmini, G.K.A. Dias, S.T. Nandasara[2017] Recognizing ancient Sinhala Inscription Characters

using Neural Network Technologies. Available at: https://ijseas.com/volume3/v3i1/ijseas20170104.pdf

4. BapuChendage, Rajivkumar Mente, Vikas Magar[2017] A Survey on Ancient Marathi Script Recognition from Stone Inscriptions. https://www.researchgate.net/publication/343808844_A_Survey_on_Ancient_

Marathi_Script_Recognition_from_Stone_Inscriptions

5.  Abhishek Tomar , Minu Choudhary , Amit Yerpude [2015] Ancient Indian Scripts Image Pre-Processing and Dimensionality Reduction for Feature Extraction and Classification: A Survey [ONLINE] Available at: http://www.ijcttjournal.org/2015/Volume21/number2/IJCTT-V21P116.pdf

6. Anush Goel, Akash Sehrawat, Ankush Patil, Prashant Chougule and Supriya Khatavkar (2018), “Raspberry Pi based reader for blind people”, International Research Journal of Engineering and Technology (IRJET 2018), vol: 5, issue: 6, pp: 1639- 1642. Available at:   

https://www.researchgate.net/publication/362491389_Raspberry_Pi_Based_Reader_for_Blind_People

7. Shuping Liu, Yantuan Xian, Huafeng Li, and Zhengtao Yu, “Text Detection in Natural Scene Images Using Morphological Component Analysis and Laplacian Dictionary”, IEEE/CAA JOURNAL OF AUTOMATICA SINICA, 2016.  Available at :https://www.ieeejas.net/article/doi/10.1109/JAS.2017.7510427

8.  Pritpal Singh, Sumit Budhiraja, “Feature Extraction and Classification Techniques in O.C.R. Systems for Handwritten Gurmukhi Script – A Survey”, International Journal of Engineering Research and Applications (IJERA). Available at :https://www.ijera.com/papers/Vol%201%20issue%204/BQ01417361739.pdf

9. Sachin S Bhat, H.V. Balachandra Achar [2015] Character recognition and Period prediction of ancient Kannada Epigraphical scripts [ONLINE] Available at: http://www.ijarcce.com/upload/2016/si/nCORETech -16/nCORETech%2024.pdf

10. Soumya A, G Hemantha Kumar,” Preprocessing of Camera Captured Inscriptions and Segmentation of Handwritten Kannada text”, International Journal of Advanced Research in Computer and Communication Engineering Vol. 3(5), May 2014.

(https://ijarcce.com/wp-content/uploads/2012/03/IJARCCE9G-s-soumya-Preprocessing-of-Camera.pdf)

11. Namrata Dave, "Segmentation Methods for Hand-Written Character Recognition", International Journal of Signal Processing, Image Processing and Pattern Recognition Vol. 8(4), 2015

12. G. Bhuvaneswari1, V. Subbiah Bharathi,” An efficient method for digital imaging of ancient stone inscriptions”, CURRENT SCIENCE, VOL. 110 (2), 25 JANUARY 2016

13. Sonika Narang, M K Jindal, Munish Kumar, “Devanagari ancient documents recognition using statistical feature extraction techniques”, Springer, 2019

14. D.A.S. Ruwanmini, K.V. Liyanage, K.G.N.D. Karunarathne, G.K.A. Dias, S.T. Nandasara, "An Architecture for an Inscription Recognition System for Sinhala Epigraphy”, International Journal of Research – GRANTHAALAYAH, Dias et. al., 4 (12): December 2016

15. Asad Iqbal Khan, AnangHudaya Muhamad Amin,” Real-Time Multi Oriented AncientScript Recognition using Single Layer Hierarchical Graph Neuron (SLHGN)”, International Conference on Applied Science and Technology (ICAST),2018

16. David Rivest-Hénault, Reza Farrahi Moghaddam, Mohamed Cheriet, “A local linear level set method for the binarization of degraded historical document images”, IJDAR (2012) 15:101– 124 DOI 10.1007/s10032-011-0157-5, 5 April 2011© Springer-Verlag 2011