Imagers with programmable, highly parallel signal processing execute computationally intensive processing steps directly on the sensor, thereby allowing early reduction of the amount of data to relevant features. For the purposes of architectural exploration during development of a novel Vision-System-on-Chip (VSoC), it has been modelled on system level. Aside from the integrated control unit with multiple independent control flows, the model also realizes digital and analogue signal processing. Due to high simulation speed and compatibility with the real system, especially regarding the programs to be executed, the resulting simulation model is very well suited for usage during application development. By providing the ability to purposefully introduce parameter deviations or defects at various points of analogue processing, it becomes possible to study them with respect to their influence on image processing algorithms executed within the VSoC

No abstract

No abstract

3D segmentation has become a widely used technique. However, automatic segmentation does not deliver high accuracy in optically dense images and manual segmentation lowers the throughput drastically. Therefore, we present a workflow for 3D segmentation being able to forecast segments based on a user-given ground truth. We provide the possibility to correct wrong forecasts and to repeatedly insert ground truth in the process. Our aim is to combine automated and manual segmentation and therefore to improve accuracy by a tunable amount of manual input.

No abstract