SearchPublisher and Institutes
You are here: Home :: Area NEM :: Engineering :: Automation
Jürgen Beyerer, Fernando Puente León
Image Optimization via Controlled Active Vision and Image Fusion
With one camera image only, it is often impossible to achieve a representation of sufficient visual quality — even if an optimal image acquisition constellation was chosen. The reason for this is the manifold interaction between the illumination, the object and the observation optics on one hand, and the limited sensing capabilities of common 2-D intensity sensors — as are used in standard cameras — to measure the information-bearing light field on the other hand. To generate images of superior quality, we record series of images with systematically varying parameters of the image acquisition system. Such series, which contain substantially more information than a single image, are then fused to resulting images featuring a quality that cannot be achieved with a single image. Based on a method to record an image with a uniform SNR at a high level, a final fusion approach is exemplarily demonstrated. Such an approach is sensible, if the image series already yields a very good insight into the quality of the future fusion result. A second example shows how — for an optically difficult biologic probe — an image series is obtained by varying the illumination direction two-dimensionally and is iteratively fused to a high-quality image result. The quality of the result of the iterative fusion determines in this case the stop of the data acquisition. In both examples, image information is fed back to the image recording system, and this way the data acquisition process is controlled with regards to a global, task-specific information gain.
at – Automatisierungstechnik, Oldenbourg Wissenschaftsverlag
Print ISSN: 0178-2312
Volume: 53, 10/2005
Pages: 493 - 502