Debarati Kundu Poster v3

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 1

3D Compression and Visual Quality Assessment

Ms. Debarati Kundu and Dr. Brian L.


Evans
debarati@utexas.edu, bevans@ece.utexas.edu

Motivation Aim 1: Static 3D Mesh Compression Aim 2: Visual Quality Assessment of Textured Meshes
Sources of 3D data Represented by triangular meshes Texture produces masking effects on geometry distortions
• PAST: Animation studios • Geometry data: Position of n vertices
• NOW: Kinect, mobile stereo, LIDAR, etc. • Connectivity data: How vertices are joined
• Texture data

Complexity: O(log n) computation and O(n4) storage Masking of surface geometry artifacts due to Masking of surface geometry artifacts due to
same spatial frequency of texture image same orientation of texture image

• Joint visual evaluation of geometry and texture leads to better


compression and mesh simplification algorithms
m = 40 m = 200 m = 500 m=1000
• Ideally work directly on surface of textured 3D objects
• No projection on several viewpoints
Manifold harmonics extend Fourier bases to surfaces
Progressive transmission of 3D data adapted to display • Useful for progressive transmission
device and network conditions in many applications • Most energy resides in lower frequency components
• 15:1 compression ratios • Some algorithms for quantifying geometric distortions correlate well with
Applications • Improvements possible due to intrinsic symmetries human perception
• Mobile, massively multiplayer and real-time gaming
• Augmented reality (e.g. 3D maps) Goal: Achieve additional 2x compression Goal: Develop perceptual quality metrics taking into account texture image
• Virtual avatars in teleconferencing
• Data storage (Google 3D warehouse) Number of polygons used to represent geometry
• Telemedicine: Transmission of 3D models of organs reduced by texture images mapped on simplified mesh Aim 3: Dynamic 3D Mesh Compression
• Scientific visualization
• Much work on compression of depth maps, which is more noise-prone
3-D Model
p=(x,y,z)
UV Map
P = (u,v)
Texture • Instead compress dynamic mesh geometry directly
• Motion paths compressed using Principal Components Analysis
• Can be combined with spectral methods for compressing spatial data

• Used for terrain rendering (OpenGL and Direct3D) Goal: Develop first perceptual measure to quantify distortions of dynamic
meshes of arbitrary connectivity without projecting to 2D

Conclusion
• Compression of 3D static and dynamic meshes result in better rendering of
• 6.7% annual growth rate to $86.1B (est.) by 2016 scenes having same visual quality at lower data rates
• Biggest growth: massively multiplayer online games • Texture compressed separately by image • Perceptual quality evaluation leads to better benchmarking of mesh
• 50% of users play exclusively on smartphones compression (JPEG, JPEG 2000) geometry and texture processing algorithms
• 66% of money on Apple & Android Apps for gaming • Joint geometry texture compression leads to better • Better understanding of psycho-visual phenomenon of shape perception
compression ratios

You might also like