Visualisation of complex flows using texture-based techniques

Queensland University of Technology

CRICOS No. 00213J

Visualisation of complex flows using texture-based techniques

D. J. Warne1,2, J. Young1, N. A. Kelson1

1High Performance Computing and Research Support, QUT2School of Electrical Engineering and Computer Science, QUT

CRICOS No. 00213Ja university for the worldreal R

Overview

• Background• Vector Field Visualisation• Traditional Techniques• Problems for Complex Flows• Advantages of Texture-Based Techniques

• Texture-Based Algorithms• Line Integral Convolution• Image Based Flow Visualisation

• Implementation and Application• Visualisation Effectiveness• Implementation Complexity• Computational Aspects

• Conclusions

CRICOS No. 00213Ja university for the worldreal R

Vector Field Visualisation

Vectors are everywhere!

“A picture says a thousand words.”

CRICOS No. 00213Ja university for the worldreal R

Traditional Techniques

We are all familiar with these:• Arrow/Quiver plots. • Streamlines/Pathlines.• Iso-surfaces.

[1] http://www.mathworks.com.au/help/matlab/ref/quiver.html

[2] http://www.mathworks.com.au/help/matlab/visualize/visualizing-vector-volume-data.html#f5-7374

[2]

[1]

CRICOS No. 00213Ja university for the worldreal R

Problems for Complex Flows

• Visual Clutter

• Choice of seed points

• Difficult to interpret time-dependent flows[3] http://rgm2.lab.nig.ac.jp/RGM2/func.php?rd_id=CircSpatial:PlotVectors

[4] J. Ma et. Al. (2011) . Streamline Selection and Viewpoint Selection via Information Channel. IEEE VisWeek Poster 2011, Providence, RI, Oct 2011.

[4]

[3]

CRICOS No. 00213Ja university for the worldreal R

Texture-Based Techniques

• Warp an image by the underlying field

• Advantages• Global/local flow regimes visible• No issues with seed points• Easily extend to capture time dependent features

CRICOS No. 00213Ja university for the worldreal R

Line-Integral Convolution (LIC)

• Applies a convolution along streamlines.

• The final image at point p is the result of a convolution of the kernel k(x) with noise along the streamline s(x,p,t) = p at x = t.

CRICOS No. 00213Ja university for the worldreal R

Line-Integral Convolution (LIC)

[4] B. Cabral, and C. Leedom (1993). Imaging vector fields using line integral convolution. SIGGRAPH 93, pp. 263-270.

CRICOS No. 00213Ja university for the worldreal R

Image Based Flow Visualisation (IBFV)

• Basic extension of LIC.

• Here, I(x,t) is now a noise image modulated in time.

• We convolve over a pathline P(x,p,t) rather than streamline.

CRICOS No. 00213Ja university for the worldreal R

Image Based Flow Visualisation (IBFV)

[5] A. Telea (2008). Data Visualization: Principles and practice. Wellesley, MA : A K Peters, Ltd, 2008.

CRICOS No. 00213Ja university for the worldreal R

Case Study: Variable-density flow through porous media

• Aquifer 600m x 200m fully saturated with fresh water.

• Sitting on top, a region of more dense salt water. • Salt water sinks into the aquifer.• Causes complex up-welling and down-welling flows.

CRICOS No. 00213Ja university for the worldreal R

Traditional Quiver PlotAnimation

CRICOS No. 00213Ja university for the worldreal R

Line Integral Convolution Image

CRICOS No. 00213Ja university for the worldreal R

Image Base Flow VisualisationAnimation

CRICOS No. 00213Ja university for the worldreal R

Visualisation Effectiveness (LIC)

LIC•Strengths

•Dense Coverage.•Spatial Correlation.•Clearly identifies extrema.

•Weaknesses•No indicators of direction.•No indicators of magnitude.•Only applicable for steady-state flows.

CRICOS No. 00213Ja university for the worldreal R

Visualisation Effectiveness (IBFV)

IBFV•Strengths

•Dense Coverage.•Spatial/Temporal Correlation.•Clearly identifies extrema.•Identifies motion of extrema.•Strong visual cues for flow direction and magnitude.

•Weaknesses•Requires animation.•Care is needed to correctly set texture speeds.

CRICOS No. 00213Ja university for the worldreal R

Implementation Comparison

LIC Algorithm1. For each pixel

1.1 Compute forward streamline.

1.2 Compute backward streamline.

1.3 Sum pixel intensities1.4 Divide by the length1.5 Assign result to output pixel.

IBFV Algorithm1. Warp mesh by field2. Render with previous

texture3. Overlay next noise

texture and blend4. Copy buffer.

CRICOS No. 00213Ja university for the worldreal R

Extensions to IBFV

• Easily extends to advection of multiple textures• Scalar data overlays. movie• Dye injects (particle traces, similar to streaklines).

movie• Jittered Grid (similar to quiver plot overlay). movie• Timelines. movie

CRICOS No. 00213Ja university for the worldreal R

Computational Aspects

• CPU based LIC can be expensive.• Need to implement interpolation.• Streamline tracing for every pixel.

• IBFV naturally implemented on GPU• Hardware handles interpolation• Convolution is written in terms of blending functions• Only mesh nodes need be intergrated

• LIC IBFV with I(x,t) = I(x)

CRICOS No. 00213Ja university for the worldreal R

Future Work

• Improve accessibility to researchers.

• Integrate into popular tools such as MATLAB.

CRICOS No. 00213Ja university for the worldreal R

Thank you!

Questions?