Document Processing Using Machine Learning

Sk Md Obaidullah, KC Santosh, Teresa Goncalves, Nibaran Das, Kaushik Roy, Sk Md Obaidullah, KC Santosh, Teresa Goncalves, Nibaran Das, Kaushik Roy

• Publisher: Chapman and Hall/CRC
• Year: 2019
• Print ISBN: 9780367218478
• eBook ISBN: 9781000739831
• Topic: Computer Science
• Subtopic: Computer Science General
• Index: Computer Science

Document Processing Using Machine Learning aims at presenting a handful of resources for students and researchers working in the document image analysis (DIA) domain using machine learning since it covers multiple document processing problems. Starting with an explanation of how Artificial Intelligence (AI) plays an important role in this domain, the book further discusses how different machine learning algorithms can be applied for classification/recognition and clustering problems regardless the type of input data: images or text.

In brief, the book offers comprehensive coverage of the most essential topics, including:

· The role of AI for document image analysis

· Optical character recognition

· Machine learning algorithms for document analysis

· Extreme learning machines and their applications

· Mathematical foundation for Web text document analysis

· Social media data analysis

· Modalities for document dataset generation

This book serves both undergraduate and graduate scholars in Computer Science/Information Technology/Electrical and Computer Engineering. Further, it is a great fit for early career research scientists and industrialists in the domain.

1

Artificial Intelligence for Document Image Analysis

Himadri Mukherjee , Payel Rakshit , Ankita Dhar , Sk Md Obaidullah , KC Santosh , Santanu Phadikar and Kaushik Roy

CONTENTS

1.1 Introduction

1.2 Optical Character Recognition

1.2.1 Dealing with Noise

1.2.2 Segmentation

1.2.3 Applications

1.2.3.1 Legal Industry

1.2.3.2 Banking

1.2.3.3 Healthcare

1.2.3.4 CAPTCHA

1.2.3.5 Automatic Number Recognition

1.2.3.6 Handwriting Recognition

1.3 Natural Language Processing

1.3.1 Tokenization

1.3.2 Stop Word Removal

1.3.3 Stemming

1.3.4 Part of Speech Tagging

1.3.5 Parsing

1.3.6 Applications

1.3.6.1 Text Summarization

1.3.6.2 Question Answering

1.3.6.3 Text Categorization

1.3.6.4 Sentiment Analysis

1.3.6.5 Word Sense Disambiguation

1.4 Conclusion

References

1.1 Introduction

There has been rapid development in technology which has aided in the digitization of documents. The number of digital documents has increased significantly over time [1, 2]. Information is now easily available on the Internet and can be distributed with ease. Such voluminous numbers of documents demand efficient processing. Digitized documents can be broadly categorized into two types, namely document images and text documents. In the case of document images, it is first essential to understand what is written. This requires optical character recognition (OCR) [3–5]. Once the characters are identified, approaches based on natural language processing (NLP) [6–8] need to be used to understand what is written. In the case of text documents, research in the fields of OCR and NLP started way back in the last century and different systems in languages like English are now commercially available [9–11], but there have not been significant developments for Indic languages. One reason for this is the complex nature of Indic scripts [12]. This is also coupled with the unavailability of standard (and free) datasets for research.

1.2 Optical Character Recognition

Optical character recognition [13, 14] refers to the task of decoding what is written in a document. It does not involve understanding the written texts, but it does involve converting a scan or a picture of a document, identifying the characters and giving the text output of the identified words and characters. The document can be either handwritten or printed. In the case of handwritten documents, there are different variations which need to be considered prior to recognition. While writing, it is often observed that the texts have disparate degrees of slants. It is very important to deal with such slants while processing the documents. A document with characters having multiple degrees of slant is presented in Figure 1.1.

FIGURE 1.1
A document depicting multiple degrees of slant for different characters.

The second important factor which needs to be tackled is the similarity between different characters. For instance, the numeral “3” is similar to “ত” in Bangla. This is illustrated in Figure 1.2. It is very important to handle such cases or else, if interpreted wrongly, the entire sentence might change.

FIGURE 1.2
Similarity between different characters in Bangla.

Another important aspect is inter-writer and intra-writer variation. It is often observed that handwritten texts show variation at the character level. That is, the same character is slightly different when written in two instances. This is known as intra-writer variation. Another variation is observed when two different writers write the same thing. The handwriting of disparate people differ from each other in most cases. This is known as inter-writer variation. Thus the system should be able to handle such differences. Inter- and intra-writer variations for a Bangla text are presented in Figures 1.3 and 1.4.

F...

Table of contents

• Cover
• Half-Title
• Title
• Copyright
• Contents
• Preface
• Editors
• Contributors
• 1. ArtificialIntelligenceforDocumentImageAnalysis
• 2. AnApproachtowardCharacterRecognitionofBanglaHandwrittenIsolatedCharacters
• 3. ArtisticMulti-CharacterScriptIdentification
• 4. AStudyontheExtremeLearningMachineandItsApplications
• 5. AGraph-BasedTextClassificationModelforWebTextDocuments
• 6. AStudyofDistanceMetricsinDocumentClassification
• 7. AStudyofProximityofDomainsforTextCategorization
• 8. SupervisedLearningforAggressionIdentificationandAuthorProfilingoverTwitterDataset
• 9. TheEffectofUsingFeaturesComputedfromGeneratedOfflineImagesforOnlineBanglaHandwrittenCharacterRecognition
• 10. HandwrittenCharacterRecognitionforPalm-LeafManuscripts
• Index