Spatiotemporal volume saliency

This paper proposes spatiotemporal volume saliency to detect and explore salient regions in time-varying volume data. Based on the center-surround hypothesis that the salient region stands out from its surroundings, we extend the spatial saliency to time domain and introduce temporal volume saliency. It is defined as a center-surround operator on Gaussian-weighted mean attribute gradient between steps in a scale-independent manner. By combing spatial saliency and temporal saliency together, our spatiotemporal volume saliency is effective in detecting changes of salient regions. We demonstrate its utility in this regard by automating transfer function design and selecting key frames for time-varying volume data.