| Author: | Wong, Kam-wah |
| Title: | Image processing for automated form reading |
| Degree: | M.Sc. |
| Year: | 2000 |
| Subject: | Image processing Optical character recognition devices Optical pattern recognition Hong Kong Polytechnic University -- Dissertations |
| Department: | Multi-disciplinary Studies Department of Electronic and Information Engineering |
| Pages: | vi, [85] leaves : ill. ; 30 cm |
| Language: | English |
| Abstract: | Forms are used extensively to collect and distributed data. It is boring and time consuming for people to read a large volume of forms. The main task of the automatic form reading is to recognize the text in binarized images and extract the useful information. In this project, we aim to develop an efficient method for image processing for automatic form reading. Binarization of a gray scale document image is the first step and most important step for automatic form reading. The performance of a form reading system is highly dependent on the performance of its binarization. In this project, a new document image preprocessing and character segmentation system was developed. The system has two main parts, image preprocessing and character segmentation. The main task of image preprocessing is to extract text fields in a form image, which mainly includes image binarization, page segmentation, and text blocks and lines extraction. The main task of character segmentation is to segment the characters in text blocks for a recognition system. A two-stage binarization scheme with feedback was developed in this project, which combines a region-based binarization technique and a neural network based binarization technique. At first, the region-based binarization is performed. After the binarized image is obtained, the horizontal Run Length Smoothing Algorithm (RLSA) followed by 8-neightbour connection checking is adopted for page segmentation. This is followed by text block/line extraction based on several simple rules. Character segmentation based on the vertical projection and a peak-to-valley function is then carried out to extract isolate characters for a recognition system. Due to the possible selection of a wrong threshold level at the region-based binarization step, the text block may become too dark or too bright, which contains connected characters or broken characters. To solve this problem, we use the neural network based binarization (four neural netwoks used with each for a type of images) to re-binarized the problematic text blocks. After these text blocks are re-binarized, character segmentation is carried out again. By using the second stage binarization, most errors in the region-based binarization can be corrected. Experimental results on a number of test images show that our two-stage binarization performs better than other single-stage binarization in terms of binarization quality and computing time. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| b15249463.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 3.54 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/3702