Abstract

Back to Publications

Author(s) Rombach, M., Saupe, D., Solzbach, U., Wollschl├Ąger, H., Just, H.
Title FOVEA - a system for dynamical comparison of compression artefacts
Abstract Although the performance of digital successors of the cinefilm is well proofed in terms of average image quality, the visual feature extraction of diagnostically significant small structures with low contrast (e.g. dissections, thrombi, stents) remains the most critical perception task during interventions. At present, there is no method accepted to quote a lossy system as to be sufficient for all those worst cases. Therefore most cardiologists demand for digital solutions as available. We tested the SIEMENS prototype of a digital solid state detector with a 1000x1000 matrix compared to a conventional image intensifier system in the same lab. We acquired pairs of loops from the same patient under identical geometry. Further test pairs came from a HICOR system and the associated cinefilm. The investigator marked the loops for the position of stents, dissections or thrombi. They were degraded by several destructive processes like down-scaling, lossy compression or storage to S-VHS tape. We used a testbed for the direct dynamical worst-case-object comparison in the running loop. The FOVEA system enhances the visual detectability of a high significant object by magnification and motion filtering. This allows to inspect the local morphology of the objects. The frame rate can be increased for better noise suppression even under magnification of the shape. The eyes of the observer can remain focused on the object. This technique allows to compare highly resolved information (e.g. from cinefilm or 1k matrix systems) on lower resolution displays (and slow internet lines). We present selected samples of the described data on the internet by ftp as an image quality test set. We invite people to use the data and post their results. In addition, we search for observers and compression developers willing to contribute to a distributed compression contest project.