Ray feature analysis for volume rendering

A major difficulty of volume rendering has been the recognition of different semantic regions which is crucial for the appropriate assignment of optical properties. Such difficulty arises from the fact that different semantic regions may share the same input value ranges. In this paper, we introduce the concept of ray-feature analysis and propose an on-the-fly state transition framework for the recognition of different semantic regions during volume rendering without the need of explicit segmentation information. In this framework, we consider the value along the path of a ray as a 1D-signal, and by making use of the feature analysis of these 1D-signals, semantic information of the current ray sample is extracted. To define the condition of state transition, we propose a method called “threshold based state transition”. Since the parameters of the threshold based state transition method is not intuitive, an automatic learning method which enables an interactive user labeling routine is proposed. Experimental results show that our proposed framework is cost effective for on-the-fly semantic region recognition, and is especially suitable for closed, mostly convex, multi-layered objects.