Form document processing based on 2-D wavelet transforms = 2차원 웨이블릿 변환에 기반한 장표문서 영상처리

다국어 초록 (Multilingual Abstract)

Form document image processing has become an increasingly essential technology in the automatic document processing. Generally, a form document processing system includes two fundamental problems: skew estimation and extraction of reference lines and items. The first problem is that the document image may appear skewed or tilted for many reasons. Therefore, the skew correction plays an important role in any automatic document analysis system. Another problem is the extraction of reference lines and items, which is the key step for form recognition and understanding. In the past years, many algorithms have been developed, however, most of them suppose that the original image is monochrome and they cannot be applied to document image with complicated background.
This thesis proposed a novel system for grey-level form document images based on the construction of 2-D wavelet transforms with adjustable rectangular supports. We first develop an approach for the skew angle estimation and skew correction. Then, a system is built for the extraction of reference lines. Finally, using the extracted reference lines and wavelet sub-images, we present a method for the extraction of item strokes. Experimental results demonstrate the effectiveness of our proposed methods.

목차 (Table of Contents)

• Abstract = ⅰ
• Contents = ⅱ
• 1 Introduction = 1
• 2 Construction of 2-D Wavelet Transforms for Form Document Processing = 6
• 2.1 Basic Concepts of Multiresolution Analysis = 6
• 2.2 2-D Orthogonal Wavelet with Square Support = 9
• 2.3 Wavelet Decomposition for Sub-Images = 12
• 2.4 Construction of Non-Orthogona1 Wavelet with Adjustable Rectangle Support = 15
• 3 Estimation of Skew Angle and Skew Correction = 18
• 3.l Related Algorithms for Skew Document Analysis = 18
• 3.2 Wavelet Decomposition of Skew Form Document Image = 22
• 3.2.1 Mathematical Description = 22
• 3.2.2 Choice of proper wavelet support = 26
• 3.2.3 Characteristics of Form Document Decomposition = 29
• 3.3 Detennination of Skew Angle and Skew Correction = 32
• 3.3.1 Binarization of sub-images = 33
• 3.3.2 Orientation of each point in the foreground = 34
• 3.3.3 Estimation of the skew angle = 35
• 3.3.4 Skew Correction using Estimated Skew Angle = 38
• 4 Extraction of Reference Lines and Items = 39
• 4.1 Overview for the Extraction of Lines and Items = 39
• 4.2 The Description of Previous wavelet-based Method = 41
• 4.3 Construction of 2-D Pseudo Wavelet = 45
• 4.4 Estimation of 2-D Wavelet Decomposition = 54
• 4.5 Horizontal and Vertical Line Extraction = 56
• 4.6 Item Extraction = 61
• 5 Experimental Results and Analysis = 65
• 5.1 Experiments = 65
• 5.2 Discussion = 72
• 6 Conclusion and Future Research = 76