Multimodal volume visualization

In "Volume Analysis Using Multimodal Surface Similarity", authors use the similarity between 2 registered volume to help users create meaningful visualization.

The similarity map, a 2D matrix, represents a measure of the similarity between every possible combination of isosurface in the volumes. For example, the value of the map at x = ( 5, 24 ) is a measure of the mutual information between isosurface of value 5 in the first volume and isosurface of value 24 in the second volume. Each cell in the map is computed by first building a distance map for each of the volumes. The distance map is a volume in which each voxel contains the shortest distance to the considered isosurface. Then, the mutual information of the 2 volumes is computed and become an entry in  the similarity map.

The similarity map can be used as an intuitive way to define a transfer function. Users can place points on the similarity map at interesting locations and assign colors and opacity to those points. The contribution of each sample in the volume to the final rendering is computed by using the colors and opacity of the nearest neighbor control points in similarity space. The figure below shows a similarity map with the control points on the left and the resulting volume rendering on the right.

Image from the paper available here

Learn from basic perception principles

The tutorial on Perception and Cognition for Visualization, Visual Data Analysis and Computer Graphics was a little shallow considering the audience. The whole morning session could have been concatenated into a half hour session without loosing any content. I still got a couple of useful insights from it to improve visualization.

A good visualization is one for which equal steps in data values produces equal perceptual steps. The next graph shows that not all visual dimensions are equal in that respect. Ideal slope is 1.

It is interesting to view the human visual system as a band-pass filter. The system is most sensitive in the a mid-range spacial frequency, but most importantly, we see from the graph that luminance is much more efficient for representing high-frequency visual details.

It is thus a good idea to use intensity to display high-frequency details in an image. On the other hand, color coding should be used for very low frequency information.

Color attracts pre-attention. It can be used by the visual system to perform an effortless search (see picture below). But color should be used with moderation. Large patches of bright color have very little effect.

The perception of a scene is highly influenced by the nature of the task performed by the viewer. The intention drives visual behavior.

Images in this post are all from the presenter's slides available here.

Re-thinking our color maps

Sunday: Tutorial on color theory, by Theresa-Marie Rhyne. Almost too basic, but interesting to get insight on how to build good color maps for visualization. The tutorial lead me to think we should pay more attention to this aspect.

First of all, I discovered the existence of a whole bunch of software tool to help build color tables:

• Adobe Kuler
• ColorBrewer
• Color Scheme designer

These tools can build color tables automatically from any picture. They also allow users to create new tables from scratch. Some have functionality to constrain user in choosing colors that follow certain basic rules of color theory. Another interesting tool is Vischeck. It takes any image as input and generates a version of that image as seen by each type of color-blind people.

Now, a couple of insights on how to mix colors.

Generally, colors that are more or less spread around a curve passing through the center of the color wheel (see picture) work well. Colors that are directly opposite in the wheel should not be used together. There should be at least a slight curvature in line passing through the color points.

Color works as an attention attractor. But the amount of a color in an image matters a lot. A small patch of red doesn't have the same effect as a large area covered with red. Bright colors should be used with moderation.


A good source of inspiration for interesting color mixes is the Color Field Painting movement.

Morris Louis, Where, an example painting
from the Color Field movement

A couple of references on color theory and art of color mixing:

• Real-world color management by Bruce Fraser and Chris Murphy
• The interaction of colors by Josef Albers
• A field guide to digital color by Maureen Stone 

Simple DTI tracts navigation

Very simple 2D interface to navigate DTI fiber tract (paper here). The method is simple. First, the tract data is analysed to cluster similar tracts together and form "bundles".

Those bundles can be represented by an average line segment that is projected in 2D. The 2D view is rendered using the Google maps API. It is then possible to select bundles from the uncluttered 2D views. All tracts that belong to the selected bundles are then highlighted in a traditional 3D rendering of all tracts (upper left corner).