CS6630 -- Project 2

Dafang

1.   PART A

Script Used:  “ PartA.tcl ”

Use information for the script:

The UI is straightforward. Switch between different objects, height/plane visualization modes, and HSV/RGB color spaces

Technical Point:

To create color entry for the RGBA color lookup table, I use a special interpolation function to make the color space as wide and diversified as possible. This design will make minor difference in RGB color more distinguishable.

      My strategy is, starting from the blue point in RGB space (0,0,1), first we fix red=0 and change color to green point ((0,1,0) in RGB space), then fix blue=0 and change color to red point ((1,0,0)) in RGB space. In this color lookup-table, blue represents the minimum value, red represents the maximum, and the green represents the middle value. This spectrum is much better than linear interpolation between red and blue, which is grey in the middle.

The visualization effect in height field:

For each object in the following pictures, the LEFT image is from the HSV color space and the RIGHT image is from the RGB color space. Also I list a table which tells the range of each component in HSV/RGB color space where the color is mapped to.

The thorax:

In this case, RGB space is better than HSV space. The reason is most data in the thorax has very similar hue value, so when I change the start/end point, all the HSV colors change at the same time, but still keep a very small distance.

     In contrast, RGB performs much better. You can see all the color spectrum is covered quite evenly. Also I try to make a large color spectrum available for visualization, as I mentioned above. We can represent small difference by colors, which is much better than represent them by saturation or intensity.

MtHood:

In this case, HSV and RGB are quite similar. In experiment I noticed that hue controls the with of cyan, blue, purple and red bands. The smaller the hue range, the narrower the width of these bands. Saturation controls the vividness of each color, and V controls the total intensity of the image.

In RGB space, the data value is directly related to a color component. The reason both methods are similar is due to the fact that the data makes a difference in hue space.

Hawaii :

In Part B,  I include a dataset for the elevation in Hawaii . The dataset, “honolulu.vtk”, is an unstructured polydata set. The elevation goes from 0 at the sea level to 1000 at the mountain peak. In the height field visualization below you can see the terrain clearly.

For clearance, the HSV plays the same as RGB space. But the visualization of RGB space is more natural, especially for the sea.

The hue range controls the highlight part of the left image: the red, pink and blue regions on the mountain peak and ridge. If we narrow the hue range, those regions will be sharp and highlighted. The saturation controls the green and cyan part of the image. The intensity controls the color of the sea. If the minimum intensity value is small, the sea will become black.

      In summary, for height field visualization, the RGB color space is better than the HSV color space. The most noticeable advantage of RGB space is its ability to visualize small differences in data. Also all the colors in the color spectrum are evenly covered by the image, where HSV model usually only uses part of the color spectrum.

Visualzation effect on plane

In each case, the left image is from HSV color space and the right image comes from the RGB color space.

Thorax:

In this case, obviously the RGB is better than HSV.

MtHood:

In this case, it’s hard to compare HSV and RGB model. The HSV model can make the very sharp boundaries between regions, and make them contrast dramatically. But it won’t change the portion of each color area much. The RGB model can’t make very sharp boundaries, but it can easily change the red, green and blue proportion. 

Hawaii :

In this case, the HSV makes a good visualization for two holes in the sea. The RGB model only uses very similar blue color to represent the sea and the holes. But again the RGB image looks more natural.　

Answer to questions in Part A & B:

1. What do you think works best, or what are the benefits of the different approaches?

From all the data samples above, color map on the height field is more illustrative than the color-mapped plane.

2. Bivariate Colormaps

Script Used: “PartC.tcl”

Use information for the script:

Each time this program only visualize one data set. In order to generate all the image, you must change the data reader in the code each time.

Technical Points

For each point in the space, I treat the low-order 8 bits and high-order 8 bits of the data as two variables x and y. Then I normalize each (x,y) dipole into [-1,1]*[-1,1] space. To build the color lookup-table, I need to map each normalized (x,y) to a unique color.

After transforming (x,y) into (magnitude, theta) coordinates, I tried to 2 colormaps.

      The first way is map theta to hue and map magnitude to saturation in HSV space.

      The second way is to map theta to hue and map magnitude to intensity in HSV space.

      It turns out that the second way is better than the first method. First, the change in intensity is more distinguishable than saturation. Second, in case that magnitude is close to zero, it doesn’t make much difference what the vector direction is.

      For the two discs below, the first one is more reasonable than the second one. Those points close to the circle center has small magnitude and small intensity and look pretty dark. But points around the center has significantly different color in the second disc, it doesn’t fit the actual situation. Also you can see the color changes radially in the first image, but it doesn’t changes so much radially in the second image. All these points to my conclusion that hue/intensity space is better than hue/saturation space for bivariate.

The Disc mapped to hue/intensity space,  “disc-8.png”:

The Disc mapped to hue/saturation space

The image generated from hue/intensity colormap for each dataset:

Mt Hood: “MtHood-8.png”

Symmetric Brain:

Non-Symmetric Brain:

Answers to Questions in Part B and CS6630 Assignment

1. What is visually wrong with the color map of the thorax data?

From the thorax data I noticed that the order of the color doesn’t coincide with the order of the value. The red color should represent the largest data value while the blue color should represent the minimum data value. But actually it’s not like what it should be. For example, on the left image of the height field thorax, the red region are right next to the blue region, which means the maximum value and minimum value are connected without any intermediate value. 

Also some place around the empty heart, different height shares the same color.

2.     What is causing this?

I guess this comes from the interpolation between the value at two points. If their difference is large but distance is small, then interpolation from neighboring points are not accurate. Also, the data value and HSV/RGB components usually doesn’t change in the same direction. This will cause some unconscious visual errors. For example, it’s hard to represent the change of data value with an increase in vividness and decrease in intensity in the same time.

To avoid these “artifacts”, we must design a colormap in which each color entry is well-refined and evenly distributed in the color spectrum, and monotonically change along one direction.

3. Are there differences between what works better for the two brain datasets?

Of course there is. For radially symmetric data, we need to map –v and v into same color. This requirement leads to a symmetric disc. But this mapping will make (x,y) and (-x, -y) have the same color in non-symmetric data, which is obvious wrong. For example, in meterological data, (20F in temperature, 100Pa in pressure) and (-20F, -100Pa) is totally different and can’t have same color.