Supervised Coloring of Volumes

We present a novel approach for assigning realistic colors to a scalar volume. Color assignment is typically done using a transfer function. However, the linear mapping between the scalar value and color produces unrealistic-looking results. In our approach, we use a reference multichannel scalar volume and its corresponding color volume to build the mapping. Using the Visible Human Male (VHM) dataset as the reference volume, we employ various machine learning techniques to find the best mapping between the VHM multichannel scalar data and the color data. We generate a feature vector for each voxel in the VHM dataset based on its 3D neighborhood. The feature vectors and their associated colors are then used to train the learning machine. After training, feature vectors from the scalar volume to be colored are fed into the learning machine to find the color. We have performed experiments on the VHM and Visible Human Female (VHF), foot and head datasets. The resultant color volumes correspond well with the full-color cross-sections of the original VH dataset.