HighQuality Spatial Interpolation of Interlaced Video

1
High-Quality Spatial Interpolationof Interlaced
Video
Alexey Lukin
Moscow State University, 2008
2
Interlaced-scan video

• Invented in 1930-ies
• Video frame is separated into 2 fields (even and
  odd raster lines)
• Improvement of motion smoothness w/o increase of
  signal bandwidth

3
Interlaced-scan video

• Invented in 1930-ies
• Video frame is separated into 2 fields (even and
  odd raster lines)
• Improvement of motion smoothness w/o increase of
  signal bandwidth

4
Deinterlacing

• Computer displays are progressive-scan ?
  deinterlacing is needed
• Simplest deinterlacing methods

? Bob (line averaging)
? Weave (field insertion)
spatial interpolation
temporal interpolation
5
Deinterlacing

• Advanced deinterlacing methods
• Motion-adaptive use Bob method in motion
  areas, use Weave method in still areas
• Motion-compensated use motion compensation to
  temporally align fields in motion-adaptive method

6
Spatial interpolation

• High-quality spatial interpolation
• Simple approaches
• Line averaging, cubic interpolation
• ELA (Edge-Directed Line Averaging)

7
Spatial interpolation

• Problem with ELA uncertain interpolation
  direction in presence of thin lines

Line averaging
ELA (5 directions)
8
Proposed method

• Aperture extension
• Allows interpolating near-horizontal edges
• Spatial averaging of derivatives
• Improves the robustness of edge sensing

9
Proposed method

• Mixing of interpolation directions
• Tolerates inaccurate detection
• of edge direction
• Expectation Maximization
• algorithm
• Re-estimation of derivatives
• from the interpolated image

10
Results
PSNR evaluation
11
Results
Visual quality
Line averaging
ELA (5-directional)
EEDI2 method (free software)
EDDI method (G. de Haan)
MSU algorithm, 2003
Proposed method
12
Results
Visual quality
Proposed method motion compensation
ELA (5-directional)
Proposed method
13
Conclusion

• The developed method has
• High visual quality and good PSNR figures
• Simple structure
• High computational complexity (optimizations
  possible)

14
Your questions
?
Thank you
http//imaging.cs.msu.ru/lukin/deinterlacing.html