薛健鑫, 蔡昆修, 何星翰

1

Introduction Work progress SR Algorithm Evaluation SR video player demo Conclusion

2

Super-resolution are techniques that improve image quality from low-resolution.

3

Multiple neighboring frames of video each provide a rich amount of information about scene details.

Super-resolution on video can analyze multiple

frames to reconstruct a enhance frame with detail

4

Retrieving image stream

From OpenCV

Super-resolution processing

Algorithm : FGSR, FRSR

GPU speedup investigation

Video player

5

Fast General Super-Resolution (FGSR)

For general video use

Fast, cheap, bad result

Fast & Robust Super-Resolution(FRSR)

Only appropriate for translational motion

Slow, better result

6

Predict HR = bicubic( Gi ) Iterative fix Predict HR :

In’ = optcial_flow( bicubic(Gi+n ) ) Predict HR += α * ( Predict HR – In’ )

*

7

8

0

50

100

150

200

250

300

350

400

450

500

64 128 256 512 1024 2048

tim

e (m

s)

Image size

Processing time

cpu

gpu

9

Bilinear interpolation FGSR

Predicted HR= Median( shift(LR0~n) ) Iterative improve Predicted HR :

Gback = FastGradientBackProject(HR);Greg = GradientRegularization(HR);HR = HR - β*(Gback + α* Greg);

10

11

Bilinear interpolation FRSR

PSNR (Peak Signal to Noise Ratio):

12

Ground truth Bilinear interpolation Bicubicinterpolation

image

PSNR(dB)

∞ 14.450087 15.79615

PSNR (Peak Signal to Noise Ratio):

13

Ground truth FGSRalgorithm

FRSRalgorithm

image

PSNR(dB)

∞ 14.642818 15.150817

14

15

Problem on super-resolution:

Optical flow

Occlusion/ disocclusion

Insufficient information from neighboring frames

Still have much space to improve

Sina Farsiu, Dirk Robinson, Michael Elad, Peyman Milanfar. Fast and Robust Multi-Frame Super-Resolution. In IEEE Transactions on Image Processing2003.

Zhongding Jiang, Tien-tsin Wong, Hujun Bao. Practical Super-Resolution from Dynamic Video Sequences. In Proc. of IEEE CVPR2003.

16

THE END

17