Representation of partially-occluded structures in regular point clouds from 3D scanner data using vision-based modulation

The reconstruction, classification and representation of objects and structures acquired using high resolution 3D scanning devices (such as medical CT or MRI) is a complex task, given the size and properties of the resulting data set. One of the challenges in dealing with such data is the identification and representation of structures partially occluded by regions with similar density values, which traditional classification methods cannot account for. We propose a simple and effective approach to this problem, which involves incorporating vision-oriented criteria into the classification process. We first reconstruct the acquired discrete data through linear texture filtering and traverse it using a volume ray caster. At each sampled point during traversal, we compute the local gradient and estimate the level of occlusion using opacity accumulation. We then modify the opacity of each sample by modulating it with coefficients derived from the previous gradient and occlusion estimations. The resulting opacity distribution makes it easy to identify partially-occluded data, which allows for the visual analysis of difficult-to-detect anatomical structures based on their level of visibility. We test our approach on CT data sets and generate images which show the projections resulting from the modified ray casting traversal loop.