Supplementary results for Neel Joshi's Dissertation, Chapter 5

Please mouse over (but don't click!) the colored underlined link to blink between our result and one of the other results or inputs.

Jump to specific results: Deblurring, Upsampling, and Denoising


Deblurring results: synthetically blurred/noised images

SIGGRAPH, noise std. dev. = 0.01

PSF (enlarged) Compare ours our result, PSNR=32.21 to:
blurred/noisy (noise sigma=0.01)
lucy richardson, PSNR=25.93
sparse prior (levin et al.), PSNR=29.44
groundtruth
blurred/noisy (noise sigma=0.01) lucy richardson, PSNR=25.93 sparse prior (levin et al.), PSNR=29.44 our result, PSNR=32.21 groundtruth

Peppers, noise std. dev. = 0.01

PSF (enlarged) Compare ours our result, PSNR=32.49 to:
blurred/noisy (noise sigma=0.01)
lucy richardson, PSNR=30.17
sparse prior (levin et al.), PSNR=32.25
groundtruth
blurred/noisy (noise sigma=0.01) lucy richardson, PSNR=30.17 sparse prior (levin et al.), PSNR=32.25 our result, PSNR=32.49 groundtruth

Three Colors Meeting at a point

PSF (enlarged) Compare ours our result, PSNR=39.83 to:
blurred/noisy (noise sigma=0.01)
lucy richardson, PSNR=28.69
sparse prior (levin et al.), PSNR=35.48
groundtruth
blurred/noisy (noise sigma=0.01) lucy richardson, PSNR=28.69 sparse prior (levin et al.), PSNR=35.48 our result, PSNR=39.83 groundtruth

Sinusoidal Pattern

PSF (enlarged) Compare ours our result, PSNR=18.22 to:
blurred/noisy (noise sigma=0.01)
lucy richardson, PSNR=16.63
sparse prior (levin et al.), PSNR=17.14
groundtruth
blurred/noisy (noise sigma=0.01) lucy richardson, PSNR=16.63 sparse prior (levin et al.), PSNR=17.14 our result, PSNR=18.22 groundtruth


Deblurring results: real images

Tour de France Map

PSF (enlarged) Compare ours to: blurred, LR, sparse blurry/noisy image Lucy Richardson sparse prior [Levin et. al] our local two color model

Dragon

PSF (enlarged) Compare ours to: blurred, LR, sparse blurry/noisy image Lucy Richardson sparse prior [Levin et. al] our local two color model

Sweater

PSF (enlarged) Compare ours to: blurred, LR, sparse blurry/noisy image Lucy Richardson sparse prior [Levin et. al] our local two color model

Fountain

PSF (enlarged) Compare ours our result to:
original blurred image
lucy richardson
sparse prior (levin et al.)
original blurred image lucy richardson sparse prior (levin et al.) our result


Upsampling results

Child's face

Quarter size image Compare ours to: bicubic, Fattal, sparse, original bicubic [Fattal 2007] sparse prior (using [Levin et. al]'s prior in our up-sampling deconvolution framework) our local two color model original image

Statue

Quarter size image Compare ours to: bicubic, Fattal, sparse bicubic [Fattal 2007] sparse prior (using [Levin et. al]'s prior in our up-sampling deconvolution framework) our local two color model


Denoising results

100075 noise level = 5%

Compare ours using two color prior, PSNR=32.34.png to:
noisy
sparse denoising, PSNR=32.35.png
0th order MRF (Liu et al.), PSNR=31.69.jpg
1st order MRF (Liu et al.), PSNR=31.68.jpg
wavelet, PSNR=31.27.jpg
groundtruth
noisy sparse denoising, PSNR=32.35.png 0th order MRF (Liu et al.), PSNR=31.69.jpg 1st order MRF (Liu et al.), PSNR=31.68.jpg wavelet, PSNR=31.27.jpg groundtruth

22013 noise level = 5%

Compare ours using two color prior, PSNR=32.57.png to:
noisy
sparse denoising, PSNR=32.49.png
0th order MRF (Liu et al.), PSNR=32.17.jpg
1st order MRF (Liu et al.), PSNR=32.33.jpg
wavelet, PSNR=31.31.jpg
groundtruth
noisy sparse denoising, PSNR=32.49.png 0th order MRF (Liu et al.), PSNR=32.17.jpg 1st order MRF (Liu et al.), PSNR=32.33.jpg wavelet, PSNR=31.31.jpg groundtruth

134052 noise level = 5%

Compare ours using two color prior, PSNR=33.02.png to:
noisy
sparse denoising, PSNR=33.01.png
0th order MRF (Liu et al.), PSNR=32.58.jpg
1st order MRF (Liu et al.), PSNR=32.88.jpg
wavelet, PSNR=32.09.jpg
groundtruth
noisy sparse denoising, PSNR=33.01.png 0th order MRF (Liu et al.), PSNR=32.58.jpg 1st order MRF (Liu et al.), PSNR=32.88.jpg wavelet, PSNR=32.09.jpg groundtruth

23084 noise level = 5%

Compare ours using two color prior, PSNR=33.10.png to:
noisy
sparse denoising, PSNR=33.00.png
0th order MRF (Liu et al.), PSNR=32.04.jpg
1st order MRF (Liu et al.), PSNR=32.64.jpg
wavelet, PSNR=32.14.jpg
groundtruth
noisy sparse denoising, PSNR=33.00.png 0th order MRF (Liu et al.), PSNR=32.04.jpg 1st order MRF (Liu et al.), PSNR=32.64.jpg wavelet, PSNR=32.14.jpg groundtruth

35008 noise level = 5%

Compare ours using two color prior, PSNR=35.69.png to:
noisy
sparse denoising, PSNR=35.89.png
0th order MRF (Liu et al.), PSNR=34.84.jpg
1st order MRF (Liu et al.), PSNR=35.97.jpg
wavelet, PSNR=35.74.jpg
groundtruth
noisy sparse denoising, PSNR=35.89.png 0th order MRF (Liu et al.), PSNR=34.84.jpg 1st order MRF (Liu et al.), PSNR=35.97.jpg wavelet, PSNR=35.74.jpg groundtruth

113009 noise level = 5%

Compare ours using two color prior, PSNR=33.55.png to:
noisy
sparse denoising, PSNR=33.61.png
0th order MRF (Liu et al.), PSNR=32.31.jpg
1st order MRF (Liu et al.), PSNR=32.61.jpg
wavelet, PSNR=32.89.jpg
groundtruth
noisy sparse denoising, PSNR=33.61.png 0th order MRF (Liu et al.), PSNR=32.31.jpg 1st order MRF (Liu et al.), PSNR=32.61.jpg wavelet, PSNR=32.89.jpg groundtruth

65010 noise level = 5%

Compare ours using two color prior, PSNR=32.42.png to:
noisy
sparse denoising, PSNR=32.26.png
0th order MRF (Liu et al.), PSNR=31.99.jpg
1st order MRF (Liu et al.), PSNR=32.18.jpg
wavelet, PSNR=30.95.jpg
groundtruth
noisy sparse denoising, PSNR=32.26.png 0th order MRF (Liu et al.), PSNR=31.99.jpg 1st order MRF (Liu et al.), PSNR=32.18.jpg wavelet, PSNR=30.95.jpg groundtruth

105053 noise level = 5%

Compare ours using two color prior, PSNR=34.57.png to:
noisy
sparse denoising, PSNR=34.79.png
0th order MRF (Liu et al.), PSNR=33.77.jpg
1st order MRF (Liu et al.), PSNR=34.02.jpg
wavelet, PSNR=34.01.jpg
groundtruth
noisy sparse denoising, PSNR=34.79.png 0th order MRF (Liu et al.), PSNR=33.77.jpg 1st order MRF (Liu et al.), PSNR=34.02.jpg wavelet, PSNR=34.01.jpg groundtruth

108073 noise level = 5%

Compare ours using two color prior, PSNR=32.65.png to:
noisy
sparse denoising, PSNR=32.49.png
0th order MRF (Liu et al.), PSNR=31.94.jpg
1st order MRF (Liu et al.), PSNR=31.98.jpg
wavelet, PSNR=31.48.jpg
groundtruth
noisy sparse denoising, PSNR=32.49.png 0th order MRF (Liu et al.), PSNR=31.94.jpg 1st order MRF (Liu et al.), PSNR=31.98.jpg wavelet, PSNR=31.48.jpg groundtruth

100075 noise level = 10%

Compare ours using two color prior, PSNR=28.76.png to:
noisy
sparse denoising, PSNR=28.50.png
0th order MRF (Liu et al.), PSNR=28.14.jpg
1st order MRF (Liu et al.), PSNR=28.96.jpg
wavelet, PSNR=28.31.jpg
groundtruth
noisy sparse denoising, PSNR=28.50.png 0th order MRF (Liu et al.), PSNR=28.14.jpg 1st order MRF (Liu et al.), PSNR=28.96.jpg wavelet, PSNR=28.31.jpg groundtruth

22013 noise level = 10%

Compare ours using two color prior, PSNR=28.02.png to:
noisy
sparse denoising, PSNR=27.52.png
0th order MRF (Liu et al.), PSNR=27.14.jpg
1st order MRF (Liu et al.), PSNR=28.84.jpg
wavelet, PSNR=27.12.jpg
groundtruth
noisy sparse denoising, PSNR=27.52.png 0th order MRF (Liu et al.), PSNR=27.14.jpg 1st order MRF (Liu et al.), PSNR=28.84.jpg wavelet, PSNR=27.12.jpg groundtruth

134052 noise level = 10%

Compare ours using two color prior, PSNR=28.97.png to:
noisy
sparse denoising, PSNR=28.14.png
0th order MRF (Liu et al.), PSNR=28.61.jpg
1st order MRF (Liu et al.), PSNR=29.55.jpg
wavelet, PSNR=28.20.jpg
groundtruth
noisy sparse denoising, PSNR=28.14.png 0th order MRF (Liu et al.), PSNR=28.61.jpg 1st order MRF (Liu et al.), PSNR=29.55.jpg wavelet, PSNR=28.20.jpg groundtruth

23084 noise level = 10%

Compare ours using two color prior, PSNR=27.96.png to:
noisy
sparse denoising, PSNR=27.18.png
0th order MRF (Liu et al.), PSNR=27.05.jpg
1st order MRF (Liu et al.), PSNR=29.23.jpg
wavelet, PSNR=27.24.jpg
groundtruth
noisy sparse denoising, PSNR=27.18.png 0th order MRF (Liu et al.), PSNR=27.05.jpg 1st order MRF (Liu et al.), PSNR=29.23.jpg wavelet, PSNR=27.24.jpg groundtruth

35008 noise level = 10%

Compare ours using two color prior, PSNR=32.27.png to:
noisy
sparse denoising, PSNR=32.15.png
0th order MRF (Liu et al.), PSNR=30.24.jpg
1st order MRF (Liu et al.), PSNR=33.27.jpg
wavelet, PSNR=31.23.jpg
groundtruth
noisy sparse denoising, PSNR=32.15.png 0th order MRF (Liu et al.), PSNR=30.24.jpg 1st order MRF (Liu et al.), PSNR=33.27.jpg wavelet, PSNR=31.23.jpg groundtruth

113009 noise level = 10%

Compare ours using two color prior, PSNR=30.05.png to:
noisy
sparse denoising, PSNR=29.87.png
0th order MRF (Liu et al.), PSNR=28.89.jpg
1st order MRF (Liu et al.), PSNR=30.19.jpg
wavelet, PSNR=29.91.jpg
groundtruth
noisy sparse denoising, PSNR=29.87.png 0th order MRF (Liu et al.), PSNR=28.89.jpg 1st order MRF (Liu et al.), PSNR=30.19.jpg wavelet, PSNR=29.91.jpg groundtruth

65010 noise level = 10%

Compare ours using two color prior, PSNR=27.63.png to:
noisy
sparse denoising, PSNR=27.05.png
0th order MRF (Liu et al.), PSNR=27.18.jpg
1st order MRF (Liu et al.), PSNR=28.41.jpg
wavelet, PSNR=26.73.jpg
groundtruth
noisy sparse denoising, PSNR=27.05.png 0th order MRF (Liu et al.), PSNR=27.18.jpg 1st order MRF (Liu et al.), PSNR=28.41.jpg wavelet, PSNR=26.73.jpg groundtruth

105053 noise level = 10%

Compare ours using two color prior, PSNR=31.59.png to:
noisy
sparse denoising, PSNR=31.72.png
0th order MRF (Liu et al.), PSNR=30.63.jpg
1st order MRF (Liu et al.), PSNR=31.95.jpg
wavelet, PSNR=31.41.jpg
groundtruth
noisy sparse denoising, PSNR=31.72.png 0th order MRF (Liu et al.), PSNR=30.63.jpg 1st order MRF (Liu et al.), PSNR=31.95.jpg wavelet, PSNR=31.41.jpg groundtruth

108073 noise level = 10%

Compare ours using two color prior, PSNR=28.73.png to:
noisy
sparse denoising, PSNR=27.92.png
0th order MRF (Liu et al.), PSNR=28.33.jpg
1st order MRF (Liu et al.), PSNR=29.21.jpg
wavelet, PSNR=27.94.jpg
groundtruth
noisy sparse denoising, PSNR=27.92.png 0th order MRF (Liu et al.), PSNR=28.33.jpg 1st order MRF (Liu et al.), PSNR=29.21.jpg wavelet, PSNR=27.94.jpg groundtruth