DAS-Discussion:Camera-Based DIA

DAS Working Subgroup Meeting: Camera-Based DIA

Authors:

• Masakazu Iwamura (Osaka Prefecture University, Japan masa@cs.osakafu-u.ac.jp )
• Elisa H. Barney Smith (Boise State University, USA EBarneySmith@boisestate.edu )

Introduction

Camera-Based Document Image Analysis (DIA) is a research field based on document images captured by cameras instead of scanners. Research work include processing and understanding document images captured by cameras and contents of the documents such as characters, layouts and so on. The research field has grown because portable cameras and mobile phones with a camera have been popularized rapidly.

The research field is characterized by using cameras. When scanners are used to capture document images, the environments to capture document images are well-controlled and we can assume the captured document images are stable. However, capturing with cameras, controlling environment is really difficult. Thus, the techniques in the research field essentially contain the problem how to control or beat the environment.

In the discussion group of Camera-Based DIA, 12 attendees of DAS 2010 from 7 countries (USA, Japan, Singapore, Germany, China, Spain and France) discussed what the open problems are and what proven problems are in the field.

Open Problems

The following three topics concerning the open problems arose during the discussion.

1 One of the topics was environments to be controlled because the research in the field essentially include it. The following items were pointed out by the attendees to be variables needing control:

• Orientation
• Uneven focal length
• Uneven size of text/effective text resolution
• Uneven color background
• Illumination
• Video with low resolution
• Perspective
• Non-planar
• Shadow exclusion

Actually it was difficult to have a consensus among the attendees as to for which algorithms were developed to the point that recognition was possible. One reason is that they have different ideas and requirements; even if some people think the problem is solved, other people sometimes do not think so; in other words, although every technique in the field resolves one of the problems above to some extent, some are satisfied by it and others are not. It was felt that the recognition task would be much easier if some guidelines were developed to help guide users in acquiring images.

2. The second topic was tasks to be solved, and applications which involve DIA in camera acquired images. In the discussion, the following items were pointed out by the attendees although there seem to be more and more tasks.

• Text Detection
• Tracking
• License Plate recognition
• Business card reading
• Historical book OCR
• Find a new application!

Text detection is an issue because in natural scenes text is so varied and often the text is significantly more artistic than text in printed documents. The last item is impressive because even if we have a lot of research work, the researchers may not think their work will directly yield good applications which can be used by ordinary people or want particular goals for new research works. In this sense, we are still waiting for killer applications.

3. The last topic was devices and related techniques to be improved for the research field. In the discussion, storage capacity and compression were pointed out. Some attendees strongly felt limits of devices. Since the techniques in the research field and such devices and related techniques have a close relationship, to watch trends of related fields seems important.

Proven Problems

Concerning the proven problems, it was also difficult to have a consensus like the open problems because the open problems and proven problems are two sides of the same coin. In such a case, good way to describe the boundary between the open problems and proven problems is to take into account the context of attendees’ experiences because whether the problem is solved or not depends on contexts. However, this time we did not have enough time to go into the details. I hope we can describe the map of the open problems and proven problems in the next discussions.

Conclusions

In this report we have presented the discussion held in subgroup meeting of Camera-Based DIA in DAS’2010 workshop. With the popularization of portable cameras, demands for Camera-Based DIA have been expanding and Camera-Based DIA is one of the main research fields in the document analysis community now. In order to encourage research in the field, the discussion was very fruitful to share attendees’ opinions and experiences. We hope we can have such opportunities in succeeding workshops and conferences.