Super-Resolution Presented By : Rashmi Pandey Guided By : Prof. Purvi Rekh
Super-Resolution
Presented By : Rashmi Pandey
Guided By : Prof. Purvi Rekh
Basic terminology
Low-Resolution (LR):
Pixel density within an image is small, therefore
offering less details.
High-Resolution (HR):
Pixel density within an image is larger, therefore
offering more details.
Super resolution (SR):
Obtaining a HR image from one or multiple LR
images
Useful in many practical cases where multiple
frames of the same scene can be obtained. It is
including video applications, medical imaging and
satellite imaging.
Synthetic zooming of region of interest (ROI) is
another important application in surveillance,
forensic, scientific, medical, and satellite imaging.
Applications of SR images
LR image Super Resolved image
(Conti......)
(Conti......)
Reduce the pixel size by increasing no. of pixels
per unit area.
Problem : Amount of light available per pixel also
decreases
Increase chip-size
Problem : Increase of capacitance leads storage problem
SR image Reconstruction
Advantage : Cost less and computationally effective
How to increase image resolution?
Basic Premise for SR
Introduction of Super Resolution
(Conti......)
Observation model
Assumed known
X
High- Resolution
Image H
H
Blur
1
N
F =I 1
F N
Geometric
Warp
D
D 1
N
Decimation
V 1
V N
Additive Noise
Y 1
Y N
Low- Resolution
Images
N1kkkkkk VXY
FHD
N1kkkkkk VXY
FHD
The Model as One Equation......
VX
V
V
V
X
Y
Y
Y
Y
N
2
1
NNN
222
111
N
2
1
H
FHD
FHD
FHD
Scheme for Super Resolution
Registration
Interpolation onto the HR grid
Deblurring and removing noise
Registration Represents the estimation of motion information
between the LR images and the reference LR
image
Interpolation on to Grid Obtain uniformly spaced HR image from
nonuniformaly spaced composite of LR
images
Deblurring and Removing Noise
Image restoration is applied to up sampled
image to remove blurring and noise
SR considering Registration Error
Blur Identification
Computationally Efficient SR Algorithm
Advanced Issues in SR
Non-uniform Interpolation
Classical method-multiple LR images
Example Based SR
SR from single image
Frequency domain approach
SR Image Reconstruction Algorithms
This algorithm performs exactly the 3 stages
presented before:
Registration, Interpolation and Restoration
Advantages
Few computational power
Real-time applications possible
Disadvantages
Only works exactly, when blur and noise
characteristics are the same for all LR images
Restoration step ignores errors caused in the
interpolation step
1.Non-uniform interpolation
Non-uniform interpolation SR reconstruction results by (a) nearest
neighbor interpolation, (b) bilinear interpolation, (c) non-uniform
interpolation using four LR images, and (d) de-bluring part (c).
(a) (b)
(c) (d)
2.Classical multi-frame based SR
Given: A set of low-quality images:
Required: Fusion of these images into a higher resolution image
How?
Actual super-resolution reconstruction result using sub pixel misalignment
Disadvantages : Limited only small increases in resolution
Algorithm uses a training set to learn the fine
details of an image at low resolution.
Maintain image database with HR/LR image pairs
Replace similar LR patches with corresponding HR
patches.
3.Example Based SR
+
LR HR
Disadvantages : Not guaranteed to provide HR image
Combine multi image SR with example based SR
Without use external source
Patch Redundancy :
Use patch redundancy in same scale to model multi
image super resolution problem
Use patch redundancy in different scales to model
example based super resolution problem
4.SR from single image
Bi-cubic interpolation Using Single Image
LR
Experimental Results
Experimental Results
Nearest Neighbor SR from Single image
LR
This algorithm is based on 3 principles:
Shifting property of the Fourier transform (FT)
Aliasing relationship between continuous FT of HR
image and the DFT of LR images (See below figure )
Band limited HR images
Advantages:
Clear demonstration of
LR and HR relationship
Capable of reducing h/w
complexity
Disadvantages:
Lack of data correlation in
frequency domain
5. Frequency Domain Approach
Conclusion
References 1. M. Elad and A. Feuer, “Restoration of Single Super-Resolution Image From Several Blurred,
Noisy and Down-Sampled Measured Images”, the IEEE Trans. on Image Processing, Vol. 6, no. 12, pp. 1646-58, December 1997.
2. M. Elad and A. Feuer, “Super-Resolution Restoration of Continuous Image Sequence - Adaptive Filtering Approach”, the IEEE Trans. on Image Processing, Vol. 8. no. 3, pp. 387-395, March 1999.
3. M. Elad and A. Feuer, “Super-Resolution reconstruction of Continuous Image Sequence”, the IEEE Trans. On Pattern Analysis and Machine Intelligence (PAMI), Vol. 21, no. 9, pp. 817-834, September 1999.
4. M. Elad and Y. Hel-Or, “A Fast Super-Resolution Reconstruction Algorithm for Pure Translational Motion and Common Space Invariant Blur”, the IEEE Trans. on Image Processing, Vol.10, No. 8, pp.1187-93, August 2001.
5. S. Farsiu, D. Robinson, M. Elad, and P. Milanfar, “Fast and Robust Multi-Frame Super-resolution”, IEEE Trans. On Image Processing, Vol. 13, No. 10, pp. 1327-1344, October 2004.
6. S. Farsiu, D. Robinson, M. Elad, and P. Milanfar, "Advanced and Challenges in Super-Resolution", the International Journal of Imaging Systems and Technology, Vol. 14, No. 2, pp. 47-57, Special Issue on high-resolution image reconstruction, August 2004.
7. S. Farsiu, M. Elad, and P. Milanfar, “Multi-Frame Demosaicing and Super-Resolution of Color Images”, IEEE Trans. on Image Processing, vol. 15, no. 1, pp. 141-159, Jan. 2006.
8. S. Farsiu, M. Elad, and P. Milanfar, "Video-to-Video Dynamic Superresolution for Grayscale and Color Sequences," EURASIP Journal of Applied Signal Processing, Special Issue on Superresolution Imaging , Volume 2006, Article ID 61859, Pages 1–15.
9. D. Glasner, S. Bagon and M. Irani, "Super-resolution from a single image," in IEEE 12th
International Conference on Computer Vision (ICCV 2009), Kyoto, Japan, Sep. 29 - Oct. 2,
2009, pp. 349-356.