Handwritten Historical Document Analysis, Recognition, and Retrieval — State of the Art and Future Trends

The following sections are included:

• Dedication
• Contents

Document image analysis or document recognition refers to the process of extracting valuable information from document images. Although a few optical character reading systems were already available in the 1970’s, the fundamental research activities on this challenging task has mainly emerged with the development of the scanner technologies in the 1980’s, which allowed affordable document image acquisition. At that time, the main applications were focused on office automation and the interpretation of printed material…

The following sections are included:

• Introduction
• Related Work
• The IAM-HistDB
  • Saint Gall Database
  • Parzival Database
  • George Washington Database
• Semi-Automatic Ground Truth Creation
• Conclusions
• References

The following sections are included:

• Introduction
• Description
  • CSG18
  • CSG863
  • CB55
• Creation
• Competition
  • Evaluation and Results
  • Discussion
• References

The following sections are included:

• Introduction
• Segmentation in Regions of Interest
• Region Description
• Typical Processing Steps
  • Binarization
  • Grouping Entities
  • Cutting
  • Labeling Data
• Layout Analysis Methods
  • Content Identification
  • Text Line Segmentation
• Open Problems
  • Semantical Analysis of the Layout
  • Reading Order
  • Rare Occurrences
• References

The following sections are included:

• Introduction
• Image Preprocessing and Feature Extraction
• Character Modeling
  • HMM Character Models
  • LSTM Character Models
• Automatic Transcription
• Extensions
• Conclusions
• References

Handwritten keyword spotting in historical documents bridges the gap between scanned documents, whose content remains inaccessible, and well processed documents on which a recognition systems can be successfully employed. In this chapter we provide an overview of state-of-the-art research into keyword spotting for handwritten documents first and then introduce a recurrent neural network-based method that is well suited for historical documents. We explain in detail the motivation, internal mechanism, its strengths and its limitations. We conclude the chapter with an overview of the most important historical databases used for keyword spotting.

The following sections are included:

• Abstract
• Introduction
• Related Work
  • Web Services in Document Image Analysis
  • Web Services in Other Fields
• DIVAServices - The RESTful Web Service Framework
• Core Interactions with DIVAservices
  • Accessing Method Information
  • Providing Data
  • Execution of a Method
• Example Use of DIVAservices
  • Upload the Original Image
  • Binarize the Image
  • Extracting Text Lines
  • Performing Optical Character Recognition (OCR)
• The Ecosystem of DIVAServices
  • DIVAServices-Spotlight
  • DIVAServices-WebInterface
  • DIVAServices-Management
• Conclusion and Future Work
• References

The following sections are included:

• Introduction
• Related Work
  • Document Segmentation and Annotation Systems
  • Human-Computer Interaction in Image Segmentation
• Document Graphs
  • Basic Definitions for Graphs
  • Graph Nodes
  • Graph Edges
  • Edge Weights
  • Graph Clustering
  • Split Layout Elements
  • Polygonal Graph Representation
  • Graph Evaluation
• Graph-User-Interaction: Scribbling
  • Scribbling as User Interaction Pattern
  • User Interaction Evaluation
• Conclusions and Outlook
• References

The following sections are included:

• Introduction
• The GRPOLY-DB Database
• Page Segmentation
  • Performance Evaluation of Page Segmentation
  • Word Segmentation
  • Document Image Segmentation Representation
• Text Recognition
  • Isolated Character Recognition
  • Text Line Recognition
• Keyword Spotting
• Conclusions
• References

Nowadays there is a great need for accessing the textual content of many vast digitized handwritten collections held by digital libraries, archives, cultural institutions, etc. The manual transcription of such huge material would be overly expensive and unaffordable in a reasonable time. On the other hand, fully automatic, or assisted handwritten text recognition technologies, respectively, still can not face this challenge with acceptable recognition accuracy or fairish processing time. The solution proposed here focuses on probabilistic indexing and search techniques under the precision-recall tradeoff paradigm. Query-by-string, word-segmentation-free, line-level keyword spotting techniques are employed to obtain the word relevance probabilities required for probabilistic indexing. Empirical results on several historical image datasets, entailing different complexity levels, confirm the viability and adequateness of the proposed indexing and search approach.

This chapter describes a complete system for the extraction of information of historical population documents. This system has been constructed as a result of several multidisciplinary projects bringing together researchers in computer vision, pattern recognition and historical demography. The chapter describes different contributions ranging from low level processing for document image enhancement, to the semantic recognition of named entities and their linkage. The proposed system gives a key role to the user, in the transcription and validation processes. User engagement experiences of crowdsourcing and gamesourcing are described. The resulting knowledge base is organized in a social network model, so the browsing of the data can be provided accordingly.

The following sections are included:

• Introduction
• Expectation Management
• Deep Learning
• The Ball-Park Principle
• Technical Realization
  • Work Flow
  • Quality and Quantity of Material
  • Industrialization and Scalability
  • Human Effort
  • Algorithms
  • Object of Recognition: Whole-Word Approaches
  • Processing Pipeline
• Performance
• Compositionality
• Conclusion
• References

Automatic text extraction from document images has been an important research area for nearly half a century. While printed text recognition has reached a level of maturity that allows commercial application for office automation, automatic transcription of handwritten documents remains a major challenge. The additional difficulty handwriting recognition is facing in comparison with printed text recognition can be explained by several reasons. An evident factor is the extreme variability of writing styles related to multiple socio-educative and cultural aspects. Additionally to these interpersonal variabilities, one can also observe significant intrapersonal variabilities related to psychological and aging factors. However, the latter being less important, writer models specifically trained for a given writer, can lead to acceptable transcription results…

The following section is included:

• Index