A Mutual Information Based Approach to Optimising View Orientation for Direct Volume Rendering

When simulating and visualising volumetric data, we often do not have the means to control the visualisation. This can be due to performance-cost or usage with an HPC where no graphical output is available. We propose a method to measure and determine a metric for the current visualisation's view, using the mutual information shared by the volumes raw simulated data and the visualisation's view achieved directly in the CUDA kernel during simulation. Only the view's rotation around a sphere is considered whilst maintaining all other degrees of freedom at a constant. A promising result is achieved showing clear/notable areas around the visualisation. We also discuss further work to improve the effectiveness of the visualisation metric.