Fraunhofer IPK

Institute for Production Systems and Design Technology

Security Technology

The ability to process visual information is a basic requirement for many automated processes. Since the early 1980s, the Security Technology department has been using innovative digital image processing and pattern recognition methods to teach machines how to »see«. The department has become internationally renowned, mainly due to a software that virtually reconstructs torn up documents.

The automation of industrial processes brought with it the need to equip technical systems with visual capabilities. A machine can only work autonomously if it can »recognize« a component, as well as its position and condition. Furthermore, machine vision opens the door to applications that would other­wise be inconceivable, such as optical inspection systems that assess structures the human eye can detect only barely or not at all. In addition to this, they operate in environments where the safety conditions mean it would be irresponsible to use human workers. Similar circumstances apply to measuring and process control systems.

Our Security Technology department is a driving force behind the development of image analysis systems. It was here where we first developed learning methods for inspecting ­material surfaces for quality assurance purposes, for controlling handling and assembly processes and monitoring systems that ensure the safety of workers in hazardous areas. Since the 1990s, we have opened up new fields by transferring production engineering solutions to non-industrial applications. Since then, the department has become an internationally recognized player in the fields of image segmentation, character and document interpretation, movement analysis, object classification, and biometric recognition systems. For example, they have developed systems for vehicle recognition and personal identification using biometric features.

Automated virtual reconstruction  

Fraunhofer IPK's most prominent application in the field of image processing is a technology that virtually reconstructs documents that have been torn up or shredded. This allows the analysis of damaged archive material and pieces of evidence. This system has been developed to allow us to read back over files shredded by East Germany's State Security Service (Stasi). However, the procedure could also be useful for a variety of other applications. The heart of the virtual reconstruction system is the ePuzzler, a software developed at IPK that uses new image processing and pattern recognition algorithms to piece together scanned in fragments of paper to form complete pages. The system uses an adaptive, non-deterministic workflow to process a wide range of characteristics, such as the contour, color, writing and lines of the fragments.

From reconstruction to »repair«

As we continue to expand on the idea of virtual reconstruction, the next logical step would be to use it on three-dimensional objects. The aim of this – especially in archeological applications – is not simply to virtually reconstruct objects, but rather to use the result of a virtual 3D reconstruction as a template for an actual reconstruction. In future, we also plan to develop tools that will help archeologists with physical reconstruction. These include a »reconstruction robot«, which could be used to help reconstruct damaged facades or wall mosaics. This will combine the expertise of both departments of the Automation Technology division.

Pattern recognition in digital worlds

Activities related to pattern recognition in digital media are currently taking us back to our safety and security engineering roots. Tools for analyzing the content of photos and films are designed to help investigators find illegal content on data carriers and the Internet. By analyzing images from ­security cameras and looking for patterns of movement that ­suggest aggression or fear, we can help to ensure safety in public spaces. The algorithms developed for this purpose are also of great interest in the media industry, for applications such as the automation of time-consuming media analyses and detecting ­plagiarism.