A Multimodal Framework for the Recognition of Ancient Tamil Handwritten Characters in Palm Manuscript Using Boolean Bitmap Pattern of Image Zoning

Transcription

1 A Multimodal Framework for the Recognition of Ancient Tamil Handwritten s in Palm Manuscript Using Boolean Bitmap Pattern of Zoning E.K.Vellingiriraj, Asst. Professor and Dr.P.Balasubramanie, Professor Department of Computer Science & Engineering, Kongu Engineering College, Perundurai, girirajek@rediffmail.com Abstract: Tamil is one of the oldest languages in the world with rich literature. In the ancient days, the writers, especially in Tamilnadu, used palm leaves to encrypt their writing. A very good example of the usage of Palm leaf manuscripts to store the history is Tamil grammar book named Tolkappiyam which was written during 4th B.C. The ancient literature includes many palm leaf manuscripts that contain Sangam works, classics, Saiva, Vaishnava and Jain works, medical works, food, astronomy & astrology, vaastu & Kaama shastra, jewellery, music, dance & drama, medicine, Siddha and so on. Over the 3, 500 Tamil manuscripts are available in Saraswathi Mahal Library located in Thanjavur, Taminadu, India. In this library, only a few palm leaf manuscripts are digitalized and many are to be digitalized so as to enable quick reference in the future. The objective of the proposed research is to develop the model that can recognize Tamil characters from palm manuscripts and convert them into text format. In the field of handwritten character recognition, image zoning is a widespread technique for feature extraction since it is rightly considered to be able to cope with handwritten pattern variability. 1. Introduction: Tamil Handwritten character recognition is one of the most difficult tasks in the pattern recognition system. There are lots of difficult things which can be solved through image processing technique: separating each character, recognizing character fonts and written styles used in different centuries. Many researchers try to apply many techniques for breaking through the complex problems of Tamil handwritten character recognition. The optical character recognition (OCR) is one of the techniques which can be defined as the process of recognizing either printed or handwritten text from document images and converting them into text format. There are many algorithms being used in the literature to perform this conversion task for specific language. In Tamil language, there are so many researchers who research using Tamil OCR but they are unable to obtain not more than 90% of accuracy. The problem is to go one step up to recognize the ancient Tamil characters. The objective of the current research is to recognize the ancient Tamil characters from old palm manuscripts by converting them into text format. 2. Related Work: According to [1], the image of palm manuscripts is used to extract a text line. This conventional text line extraction can be roughly classified into four categories: i. Projection based methods ii. Hough transform based methods iii. Bottom up grouping methods iv. segmentation based methods Most of works based on text line segmentation can be roughly categorized as bottom-up or top-down approaches. In the topdown methodology, a document page is first segmented into zones, and a zone is then segmented into lines, and so on. Projection based methods is one of the most successful top-down algorithms for printed documents and it can be applied on handwritings only if gaps between two neighboring handwritten lines are sufficient. Projection based methods allow the efficient extraction of text lines [2], [3]. We use this method for palm manuscripts. 139

2 3. Methodology: Here, all details of the proposed system design are given. First, the overall framework of the ancient Tamil handwritten character recognition system is given. 3.1 System Architecture Overview: Fig. 1: Framework of the Ancient Tamil Handwritten character recognition in Palm manuscripts 3.2 System Structure Chart: Based on the system framework in the previous section, the Tamil palm leaf image is converted into Tamil text format. This framework includes, i) scanning ii) preprocessing iii) Feature extraction iv) recognition v) Text conversion. Tamil Palm Leaf Recognition Scanning Preprocessing Feature Extraction Recognition Display Result Cropping Resizing Thicken Binarization to Boolean Matrix Grouping Pattern Matching Text Conversion Fig 2: Structure chart of Ancient Tamil handwritten characters recognition by image zoning using the Boolean matrix i. Scanning: In the first stage, the Tamil palm leaf manuscripts belonging to different centuries would be collected from various places in Tamilnadu. These manuscripts are scanned by 4800 dpi scanner and stored in Jpeg format. ii. Preprocessing: In the image preprocessing module, the proposed system would prepare a palm manuscript handwritten character image for the feature extraction module. This stage consists of five sub-processes: a) image cropping b) segmentation c) image re-sizing d) image thickening and e) binarization. Each of these subprocesses are given below: 140

3 a) cropping: This process involves the cropping of each word. The scanned palm leaf image would have white space. Using this white space, words are cropped. b) Segmentation: There are three types of segmentation. Line segmentation, word segmentation and character segmentation. Researchers used various techniques for segmentation like threshold techniques, region based method, edge based method, graph based methods, clustering methods, compression based methods, histogram methods, watershed transformation and model based methods. Here we will take the edge detection method to segment the characters. The edge detection is the name for a set of mathematical methods which aim at identifying points in a digital image at which the image brightness changes sharply or, more formally has discontinuities. Marr-Hildreth algorithm is one of the edge detection, which has a method of detecting edges in digital images. c) re-sizing: Each segmented character is in different size. So, it is necessary to change all the characters into equal size. In the proposed method, the character image is re-sized into 100X100 pixels. 100 X 100 Pixels d) thickening: Each darkened pixel of the re-sized character is thickened through darkening the nearest pixels. Using the nearest algorithm, a thin character is changed into thicker character through darkening the color of the nearest pixel for a target range. e) binarization: Each character is stored in Boolean matrix in either 0 s or 1 s. Using the image zoning technique [5], all the dark pixels are stored in 1 s and light pixel in 0 s. iii. Feature extraction: This feature extraction module extracts the basic components of Tamil characters. There are three sub modules a) conversion into Boolean Matrix b) Grouping and c) Pattern Matching the details of which are given below: a) conversion into Boolean Matrix: Each character from palm leaf is converted into Boolean matrix. Similarly, all the actual Tamil character sets (Fig 4) are also converted into the Boolean matrix because palm 141

4 leaf manuscripts were written in hand and might be unreadable to different people and it is necessary to make them readable. Fig 4: The Tamil character sets b) Grouping: Palm manuscripts were used in different centuries and the words appear in them are of different styles and strokes (Fig 5). Each individual script is stored in Boolean matrix. Fig 5. Tamil Scripts in Different Centuries c) Pattern Matching: The stored original text Boolean matrix would be compared with the new preprocessing Tamil character using a range. 4. Recognition: Pattern matching identifies each similar character in the predefined Boolean matrix. Then, the matching Boolean matrix is converted into equal Unicode Tamil fonts. 4. Modeling: A character model is a record of all the characters set that are of equal Boolean matrix. The combination of the two Boolean matrix is also equal to the single character. For example: The equivalent Boolean matrix is given below: + = 142

5 A simple algorithm that is can be used for character matching is as follows: M is the Boolean matrix one by one in all character sets in Tamil scripts ( set 1 to 67). N is the Boolean matrix for current handwritten character in palm manuscripts. 1. Predefine M in all Boolean matrix 2. Check the current character Boolean matrix N with M 3. If the range of both Boolean matrix is set to 1, the characters are identified ( set no 1 to 55) 4. If not, check the part of the Boolean matrix with the entire predefined matrix ( set no 1 to 30) If Boolean matrix is matches, then check all the other parts of the Boolean matrix ( set 56, 57, 58, 63, 64, 65) 5. Conclusion: In this paper, we have proposed a simple method for converting ancient Tamil handwritten scripts into text format. There are thousands of Tamil palm manuscripts that are yet to be digitalized. The aim of this paper is to convert the palm manuscript image into digitized text format. However, our method has some difficulties in handling cases such as cursive Tamil script, merging of two Boolean matrixes, and a hole in palm manuscript image. These are only some basic issues which can be overcome through future extension of character recognition. References: Hyung Il Koo and Nam Ik Cho, Text Line Extraction Chinese Documents Based on an Energy Minimization Framework, IEEE Trans. On Processing, Vol.21, no.3, pp , Mar G. Nagy, S. Seth, and M. Viswanathan, A Prototype Document Analysis System for Technical Journals, Computer, vol. 25, no. 7, pp , Jul F. Shafait, D. Keysers, and T. M. Breuel, Performance Evaluation and Benchmarking of Six-Page Segmentation Algorithms, IEEE Trans. Pattern Anal. Mach. Intell., vol. 30, no. 6, pp , Jun Y.Liang, M.C.Fairhurst & R.M.Guest, A Synthesisd Word Approach to Word Retrieval in Handwritten Documents, Elsevier Pattern Recognition, Vol.45, PP , June Giuseppe Pirlo, Donato Impedovo, Adaptive Membership Functions for Handwritten Recognition by Voronoi-Based Zoning, IEEE Trans on Processing, Vol 21, No 9, PP , Sep Chomtip Pornpanomchai, Verachag Wongsawangtham, Satheanpong Jeungudomporn, and Nannaphat Chatsumpun, Thai Handwritten Recognition by Genetic Algorithm (THCRGA), IACSIT Journal of Engineering and Technology, Vol 3, No 2, Apr Qiu-Fend Wang, Fei Yin, and Cheng-Lin Liu, Handwritten Chinese Text Recognition by Integrating Multiple Contexts, IEEE Trans on Pattern Analysis and Machine Intelligence, Vol 34, No 8, Aug Tiji M Jose and Amitabh Wahi, Recognition of Tamil Handwritten s using Daubechies Wavelet Transforms and Feed-forward Back Propagation Network, IJCA, Vol 64, No 8, PP , Feb Jin Chen, Daniel Lopresti, Model Based Ruling Line Detection in Noisy Handwritten Documents, Pattern Recognition Letters, Elsevier, A Bharath and Sriganesh Madhvanath, HMM-Based Lexicon-Driven and Lexicon-Free Word Recognition for Online Handwritten Indic Scripts, IEEE Trans on Pattern Analysis and Machine Intelligence, Vol 34, No 4, Apr