Speaker: Manuel Hochmayr (ICGA)
Parameters and the process of setting them play a major role in the world of computer based visualisation algorithms. Finding suitable parameter settings can take most of the time in a visualisation process.
For this master thesis a prototype was developed which guides the user through a semiautomatic process of adjusting parameter settings, which finally results in the desired visualization of a scientific volume. The main strategy of the thesis is to present pre-rendered images of a volume with different parameter settings to the users. The images that are closest to the target visualisation can be selected and new images, similar to these are shown or new ones depending on the before chosen ones are generated.
We make use of the concept of high-level parameters. In order to increase and decrease the influence of the high-level parameters a new approach is tried by using an arbitrary size parameter settings combination vector, for each high-level parameter, which is created automatically and used to render images according to the actual setting entry.
k-means++ or DBScan is used then, with an automatic method to choose the distance parameter, to cluster the images by their similarity and only the cluster centres are then presented to the user in the spread sheet interface for exploration. As a second new approach local linear embedding is used as similarity measurement for k-means++ and DBScan.
By combining all these techniques we show that is possible to explore many different parameter settings for the high-level parameter in very short time.