Urdu Script Recognition - Assignment Example

Urdu OCR 1- Executive Summery This research presents the detailed analysis of the digital image processing techniques used in Urdu OCR. In this research I will discuss about the optical character recognition. This technology is really useful in present information technology structure. This report will present a detailed overview of paper “Recognition of printed Urdu script” presented by U. Pal and Anirban Sarkar. Here I will discuss different algorithms and techniques regarding the Urdu Optical Character Recognition presented for the enhanced character recognition. 2- Introduction Optical Character Recognition (OCR) refers to the areas or branch of computer science that engages reading of text from paper as well as translating the images into a structure that computer is able to recognize (for instance converting into ASCII codes). An Optical Character Recognition system allows us to take a magazine or book or article, feed it straightly into an electronic computer data file, moreover edit the file by means of a word processor (webopedia, 2009). Urdu language is similar to Arabic, which is used widely in different countries. There is no such work was done previously. This research has offered a better reorganization of Urdu script. Pal & Sarkar (2003) also developed a prototype of the system that has attained 97.8% character level accuracy on average (Pal & Sarkar, 2003). 3- Main Structure of OCR All Optical Character Recognition systems comprise an optical scanner intended for extracting or reading text, as well as really complicated/sophisticated software intended for analyzing images. The majority Optical Character Recognition systems utilize a blend of hardware (particular circuit boards) as well as software to identify characters, while a number of low-priced systems perform it completely through software. Superior Optical Character Recognition systems are able to read text in huge variety of fonts; however they still have trouble through handwritten text (webopedia, 2009). The power and effectiveness of Optical Character Recognition systems is huge since they facilitate users to control the power of computers systems to review printed documents. Optical Character Recognition is previously being utilized extensively in the official profession, education, research, and print media (webopedia, 2009). But there is less amount of work done on recognition of other languages (i.e Arabic, Hindi, Urdu). 4- Urdu Script overview and difficulties with OCR implementation The Urdu script is a complex language script. The total number of alphabets in Urdu is 39. In this language we have 10 numerals characters. The main difficulty of this language is the compound characters those are formed through the combination of some characters. The main difficulty in Optical Character Recognition system is difficulty to detect from these compound shapes. The main characteristic of the Urdu script is the similarity of different shapes in the overall alphabets of script. This aspect is also a main difficulty in the recognition of Urdu script (Pal & Sarkar, 2003). The character recognition of English language is very simple because space between characters can be recognized there. The image given below shows the alphabets in detail: Figure 1 Urdu Alphabets: Source (Pal & Sarkar, 2003) 5- Proposed System Pal & Sarkar (2003) has proposed Optical Character Recognition system that utilizes the technique for recognition of individual characters. This process recognizes the Urdu script by means of a combination of contour, topological, and a new concept in this area “water reservoir” based features of Urdu characters. The implemented techniques for character recognition techniques are robust and simple (Pal & Sarkar, 2003). Pal & Sarkar (2003) has utlized the special technique of “segmentation”. This character segmentation technique offers a great advantage regarding the improvement for handling a huge variety of Urdu characters that occur frequently in images taken from inferior quality source documents. This system can perform effectively if fine-tuned for the wider variety of images enclosing characters in varied sizes and fonts (Pal & Sarkar, 2003). This technique is useful only for few alphabets. This technique does not work for all the alphabets. 6- OCR Algorithms This section will provide an overview of the different algorithms used in this research. 6.1- Water Reservoir Principle Pal & Sarkar (2003) have used the water reservoir principle for their Optical Character Recognition. In this technique the water is poured from one side of an element, the opening areas of the element where water will be stocked up are measured as reservoirs. Through bottom or top reservoirs we denote the reservoirs attained when water is poured as of bottom or top of the element. Also if water is poured in the element from right (left) side, the cavity areas of the components where water will be accumulated are known as right (left) reservoirs. This technique helps in recognizing different alphabets of Urdu. The entire reservoirs taken from a way of an element are not taken for future processing. Figure 2 shows water reservoir technique: Figure 2Water Reservoir, source: (Pal & Sarkar, 2003) 6.2- Skew detection and correction The conversion of image in digital format involves the utilization of histogram that is foundational upon the thresholding technique. Removal of pixels is decided on the basis of that threshold. In Skew detection technique we demonstrate object pixels through 1 and background or white pixels through 0. The 2- color based image normally demonstrates projections as well as dents in the characters and isolated object pixels in excess of the background, that are refined through a logical smoothing technique. Normal utilization of the scanner can direct to skew in the document image. This technique is useful for removing noise and unwanted pixels from the scanned image. The removal of noise and unwanted pixel is decided on the basis of threshold value. Normally, we can set any value which is between 0 and 1. In this technique skew angle is the angle that the document text line of the document image creates by means of the horizontal way. Skew improvement can be attained through initially approximating the skew angle, moreover rotating the image through the skew angle in the differing way (Pal & Sarkar, 2003). Figure 3Process of Skew detection, Source (Pal & Sarkar, 2003) 7- Detection Process This section is about the elaboration of main detection process: 7.1- Line and character segmentation The Optical Character Recognition system presented by Pal & Sarkar (2003) repeatedly perceives individual text lines as well as then sections the characters in every line. We do not fragment words as of a line intended for the detection reason. The lines of a text chunks are divided through discovering the valleys of the protuberance profile assessed through counting the amount of black pixels in every row. The channel among two successive peaks in this profile indicates the edge among two text lines. A text line is able to be divorced among two successive boundary lines (Pal & Sarkar, 2003). Figure 4lines detection, source (Pal & Sarkar, 2003) 7.2- Feature selection For the early categorization of characters, we judge contour features, topological features and features attained from then idea of water reservoirs. The topological features utilized in existence of holes. Contour characteristics comprise characteristics of diverse profiles acquired from a segment of character’s contour. The major water reservoir is foundational upon features employed in the detection scheme (Pal & Sarkar, 2003). 7.3- Character recognition Pal & Sarkar (2003) establlisged technique detects the Urdu characters in two stages. In the initial stage, the Urdu script characters are clustered into small subsets through a feature based tree segmentation. In the subsequent phase, we utilize additional sophisticated features to identify comparable characters attached to leaf nodes of the categorization tree. (Pal & Sarkar, 2003). 8- Suggestions I think the paper is excellent. But I would like to give some suggestions or ideas regarding this paper. Skeletonization is the method of peeling off of a pattern as many pixels as possible without modifying the actual form of the pattern. In more simple words, after pixels have been removed, the pattern should not change its meanings. The resultant skeleton must have the following characteristics: (Azar, 1997) as thin as possible connected centered Figure 5Image before applying hilditch algorithm Figure 6Image after applying hilditch algorithm Images source: (Azar, 1997) If we apply Hilditch algorithm on Urdu images then recognition will be uncomplicated, because it will be easy to separate and recognize single pixel text instead of bold text. 9- Conclusion Pal & Sarkar (2003) proposed OCR system for different printed Urdu documents. This system recognizes individual text lines by means of a correctness of 98.3 percent. The character division/segmentation correctness of the system is 96.9 percent. The majority segmentation faults were reasoned through the touching as well as compound characters. Occasionally a number of errors were reasoned for overlapping also. This report has presented an overview of the Pal & Sarkar (2003) proposed OCR system. I have outlined the main operational steps and working structure. I have also proposed some ideas to improve this technique. References Azar, D. (1997). Hilditch's Algorithm for Skeletonization . Retrieved October 10, 2009, from McGill University: http://cgm.cs.mcgill.ca/~godfried/teaching/projects97/azar/skeleton.html Pal, U., & Sarkar, A. (2003). Recognition of Printed Urdu Script. IEEE- Proceedings of the Seventh International Conference on Document Analysis and Recognition (ICDAR 2003) . webopedia. (2009). optical character recognition . Retrieved 10 08, 2009, from http://www.webopedia.com/TERM/O/optical_character_recognition.html Read More