英语原文共 6 页，剩余内容已隐藏，支付完成后下载完整资料

A Complete Workflow for Development of Bangla OCR

ABSTRACT

Developing a Bangla OCR requires bunch of algorithm and methods. There were many effort went on for developing a Bangla OCR. But all of them failed to provide an error free Bangla OCR. Each of them has some lacking. We discussed about the problem scope of currently existing Bangla OCRrsquo;s. In this paper, we present the basic steps required for developing a Bangla OCR and a complete workflow for development of a Bangla OCR with mentioning all the possible algorithms required.

Keywords

OCR, Bangla OCR, Bangla Font, Matra, Preprocessing, Binarization, Classification, Segmentation, Page Layout analysis, Tesseract.

INTRODUCTION

OCR means Optical Character Recognition, which is the mechanical or electronic transformation of scanned images of handwritten, typewritten or printed text into machine-encoded text.

OCR has emerged a major research field for its usefulness since 1950. Bangla is ranked 5th as speaking language in the world. With the digitization of every field, it is now necessary to digitized huge volume of old Bangla book by using an efficient Bangla OCR. However, till today there is no such OCR for Bangla is developed. The existing Bangla OCRs could not fulfill the desired result. From 80‟s Bangla OCR development has started and now for its necessity this field becomes a major research area today. For further progression, synchronization of the total system is required. Here we present a total overview of Bangla OCR and its existing challenge is to know the development procedure and to estimate what more requires to do to create a complete Bangla OCR.

RELATED WORKS

Though Bangla OCR is not a recent work, but there are very few mentionable works in this field. BOCRA and Apona-Pathak‟ was made publicly in 2006 1. But they are not open source. The Center for Research on Bangla Language Processing (CRBLP) released BanglaOCR – the first open source OCR software for Bangla – in 2007 2. BanglaOCR is a complete OCR framework, and has a recognition rate of up to 98% (in limited domains) but it also have many limitations.

STEPS OF BANGLA OCR

Although OCR system can be develop for different purposes, for different languages, an OCR system contains some basic steps. Figure-2 describes the basic steps of an OCR. A basic OCR system has the following particular processing steps:

1. Scanning.

2. Preprocessing.

3. Feature extraction or pattern recognition.

4. Recognition using one or more classifier.

5. Contextual verification or post processing

Scanning

To extract characters from scanned images it is necessary to convert the image into proper digital image. This process is called text digitization. The process of text digitization can be performed either by a Flat-bed scanner or a hand-held scanner. Hand held scanner typically has a low resolution range. Appropriate resolution level typically 300-1000 dots per inch for better accuracy of text extraction [1].

Preprocessing

Preprocessing consists of number of preliminary processing steps to make the raw data usable for the recognizer. The typical preprocessing steps included the following process:

1. Binarization

2. Noise Detection amp; Reduction

3. Skew detection amp; correction

4. Page layout analysis

5. Segmentation

Binarization methods

Binarization is a technique by which the gray scale images are converted to binary images. Some binarization methods are given below:

Global Fixed Threshold: The algorithm chooses a fixed intensity threshold value I. If the intensity value of any pixel of an input is more than I, the pixel is set to white otherwise it is black. If the source is a color image, it first has to be converted to grey level using the standard conversion [2].

Otsu Global Algorithm: This method is both simple and effective. The algorithm assumes that the image to be threshold contains two classes of pixels and calculates the optimum threshold separating those two classes so that their combined spread (intra-class variation) is minimal .

Niblackrsquo;s Algorithm: Niblacrsquo;s algorithm calculates a pixel wise threshold by sliding a rectangular window over the grey level image. The threshold is computed by using the mean and standard deviation, of all the pixels in the window[2].

Adaptive Niblackrsquo;s Algorithm: In archive document processing, it is difficult to identify suitable sliding window size SW and constant k values for all images, as the character size of both frame and stroke may vary image by image. Improper choice of SW and k values results in poor binarization. Modified Niblacrsquo;s algorithm allows automatically chosen values for k and SW, which is called adaptive Niblacrsquo;s algorithm[2].

Sauvolarsquo;s Algorithm: Sauvolarsquo;s algorithm is a modification of Niblacrsquo;s which is claimed to give improved performance on documents in which the background contains light texture, big variations and uneven illumination. In this algorithm, a threshold is computed with the dynamic range of the standard deviation[2].

Noise detectionamp;correction methods

Noise can be produced during the scanning section automatically. Two types of noises are common. They are background noise and salt amp; paper noise. Complex script like Bangla, we cannot eliminate wide pixels from upper or lower portion of a character because it may not only eliminate noise but also the difference between two characters like e and i and for some other characters[3]. Some noise detection methods are given below:

T

剩余内容已隐藏，支付完成后下载完整资料

资料编号：[148415]，资料为PDF文档或Word文档，PDF文档可免费转换为Word