2D to 3D conversion at CRC: A visual perception approach.
1. 2D to 3D conversion at CRC
A visual perception approach
Carlos Vazquez and Wa James Tam
3D-TV Research
Advanced Video Systems
2. 2D to S3D video conversion
Why? Where/When? How?
•Many S3D cinema screens available •Post-production: For cinema and •Fully manual
•Introduction of 3D-enabled TV sets TV production •Semi-automatic
•Large amount of 2D content •Broadcasting: For live content and •Automatic off-line
available legacy material •Automatic in real-time
•Limited content in original •TV set, STB: Legacy content from
stereoscopic format viewer library or 2D program
•Still difficult to capture
stereoscopic content
June 4, 2012 Advanced Video Systems -- 3D-TV Research 2
3. Real-time vs. off-line conversion
On-line 2D-to-3D conversion
Fully automatic
Artefacts are not removed
Speed of conversion is critical
Non-veridical depth information frequently
used
Off-line 2D-to-3D conversion
Automatic, semi-automatic or manual
Automatically generated content can be
corrected
Normally human-assisted (post production)
High quality and high cost
June 4, 2012 Advanced Video Systems -- 3D-TV Research 3
4. 2D to S3D conversion: General framework
Strategy:
Define depth structure and
render new virtual view
Input:
2D video sequence
Output:
Stereoscopic/multi-view
video sequence
June 4, 2012 Advanced Video Systems -- 3D-TV Research 4
5. How to get the depth
Manually assignment of
• Human interpretation of the scene
depth by operator analysis of translates into depth information
the scene
Extracting depth from • Depth from static images
monocular cues in images • Depth from motion related cues
Assigning of depth from • Apply a know depth model
automatic analysis of the • Use surrogates for the depth
scene information
June 4, 2012 Advanced Video Systems -- 3D-TV Research 5
6. Automatic generation of depth information
Two approaches are used:
Extract depth information from pictorial
depth cues in images
Assign depth based on general
assumptions about scenes
Pictorial depth cues:
Size, texture density, occlusion,
atmospheric haze, shadows
Perspective
Focus/defocus
General assumptions
Scene depth models Single image with several pictorial
Surrogate depth based on colour and depth cues
shadings
June 4, 2012 Advanced Video Systems -- 3D-TV Research 6
7.
8. Visual perception: An active process
Content dependency:
Aoccdrnig to rscheearch at an Elingsh uinervtisy,
Past experience help in interpretation of visual content
it deosn't mttaer in waht oredr the ltteers in a
Auto-correction:
wrod are, olny taht the frist and lsat ltteres are
Fill-in missing information or minimize conflicting information
at the rghit pcleas. The rset can be a toatl mses
and you can sitll raed it wouthit a porbelm. Tihs
is bcuseae we do not raed ervey lteter by ilstef,
but the wrod as a wlohe.
http://kybele.psych.cornell.edu/~edelman/
June 4, 2012 Advanced Video Systems -- 3D-TV Research 8
10. Surrogate depth maps
Motivation
Shadows offer relative depth
inside objects
Small image features (rain drops,
snow flakes, dust) have depth
Good separation between objects
in the scene
Green and blue go back (Trees
and sky)
Red goes to the front (Skin tones)
June 4, 2012 Advanced Video Systems -- 3D-TV Research 10
11. CRC-i3D: Real-time 2D to 3D video conversion
Surrogate Depth Maps:
Single Cue – using the Chroma component of the
images.
Minimal adjustments to original depth map
Rendering:
Real-Time and no buffering required.
Avoids disocclusion by smoothing depth map.
Manual control of depth to ensure comfort
according to viewing conditions.
Advantages:
Very low processing requirements.
No storage requirements.
This all translates into less processing, heat and
power requirements.
June 4, 2012 Advanced Video Systems -- 3D-TV Research 11
12. Off-line conversion: depth-based approach
Depth generation Depth map editing New image generation
Generate Select Edit depth Render Fill-in
2D depth regions values image holes
S3D
Generate initial depth: Automatic operation
Edit depth map: Manual or semi-automatic
Select objects: Semi-automatic operation
Change depth of objects: Semi-automatic
Render new view: Automatic operation
Fill-in holes: Automatic operation with human
supervision.
June 4, 2012 Advanced Video Systems -- 3D-TV Research 12
13. Depth-based approach: Main characteristics
Intermediate depth master: Can be used to render the
stereoscopic content for different targets
Depth relations are easy to evaluate
Depth is continuous: Slow variations in depth are easily
represented
Depth information is represented as an image: Easy to
store and transmit
June 4, 2012 Advanced Video Systems -- 3D-TV Research 13