### Table 1. Super-resolution output of the MAP-B estimator.

"... In PAGE 6: ... 3 shows a full frame extracted from a video showing text of various sizes from the Gettysburg Address. Table1 shows the results of a test in which the super-resolution process uses up to 32 input images of various-size text. Fig.... ..."

### Table 5: The results of the super-resolution image reconstruction.

"... In PAGE 17: ...(0; 0); (0; 2); (1; 1); (1; 3); (2; 0); (2; 2); (3; 1); (3; 3)g, see Figure 5. As in x4.2.2, the tensor product of the lowpass lter m in Example 2 is used to generate the low-resolution images, and white noises at SNR = 40dB are added. Table5 shows the results of the least squares model, and Algorithms 2 and 3 with symmetric boundary conditions. The optimal in Step 2 and Step 4 are 0:0121 and 0:0105 for the least squares method and 0:0170 and 0:0161 for Algorithm 2 respectively.... ..."

### Table 5: The results of the super-resolution image reconstruction.

"... In PAGE 17: ...(0; 0); (0; 2); (1; 1); (1; 3); (2; 0); (2; 2); (3; 1); (3; 3)g, see Figure 5. As in x4.2.2, the tensor product of the lowpass lter m in Example 2 is used to generate the low-resolution images, and white noises at SNR = 40dB are added. Table5 shows the results of the least squares model, and Algorithms 2 and 3 with symmetric boundary conditions. The optimal in Step 2 and Step 4 are 0:0121 and 0:0105 for the least squares method and 0:0170 and 0:0161 for Algorithm 2 respectively.... ..."

### Table 1: MSE comparison for the high resolution Vase and Jodu images and depth map obtained using bi-cubic interpolation and our super-resolution approach with an upsam- pling factor of 2 with different source positions. The (DEPTH) row in the table gives the MSE for the depth field.

"... In PAGE 8: ... For quantitative comparison, we use mean square error (MSE) as a figure of merit. Table1 shows the MSE comparison for the super-resolved image and the depth map (for both Vase and Jodu images) and the case when interpolated values of the surface gradients and albedo are used for reconstruction of the up-sampled depth and intensity map. Al- though, not much difference can be seen in the high resolution images reconstructed using the two methods, the MSE values clearly show that the high resolution images obtained using our graph cuts based approach are much better than those obtained using bi-cubic interpolation.... ..."

### Table 4: Contrast and Image mottle enhancement from the original input image and for each deconvolu- tion method (enhancement expressed in percentage). Respectively; the Van-Cittert (VC), the Landweber (LW), the Richardson-Lucy (RL), the Super Resolution (SR), the Molina (MO), the IBD, the Biggs-Lucy (BL), the NAS-RIF and nally, the You-Kaveh apos;s (YK) algorithm.

"... In PAGE 16: ... Due to the di erence of thickness between the cross-sectional slices of the real and segmented phantom, these abovementioned measures are estimated on the whole 3D phantom after this one has been registered [22] on the ground truth of the segmented phantom volume (see Figure 10 where some consecutive slices of the segmented phantom are shown). Table4 gives the contrast and image mottle for each deconvolution technique applied on this SPECT phantom. Amongst the supervised deconvolution schemes, the Landweber apos;s algorithm allows to increase sig- ni cantly the contrast of the image but at cost of an unacceptable increase of the mottle of the image (+33:0% of mottle).... ..."

### Table 4: The numerical (dnum) estimate of the refocusing resolution for time reversal through random media and comparison with the deterministic case.

2001

"... In PAGE 16: ... For the three other snapshots we a use random media with the same correlation length but with different variance for the fluctuations. The characteristics of the different random media are given in Table4 . We also give in this table the maximum contrast for each medium.... In PAGE 18: ... We calculate the refocusing resolution (dnum) as before by a least squares fit of the curvature at the peak amplitude (in the cross-range direction). The results are given in Table4 , where in the last column we compute the enhancement in the refocusing resolution in the random medium by comparing it to that obtained in the deterministic case. This is the super-resolution effect caused by multipathing.... In PAGE 18: ...Table 4: The numerical (dnum) estimate of the refocusing resolution for time reversal through random media and comparison with the deterministic case. The results shown in Table4 demonstrate quantitatively the super-resolution phenomenon: in media with random heterogeneities the refocusing resolution beats the diffraction limit, which is the refocusing resolution in the homogeneous medium. We see clearly that better resolution is obtained as the standard deviation of the fluctuations in the random media increases.... ..."

Cited by 5

### Table II. Averaged PSNR value in Db measured over a test pool of digital images in case of super resolution.

### Table 4: MPEG videos used in experiment

1999

"... In PAGE 20: ... Thus, the total running time for reconstructing the motion compensated (MC) blocks is determined by the number of the macroblocks with motion vectors in the P and B frames. Generally, the running time can be roughly estimated by the frame size, number of frames, and percentage of P , B frames, which are listed in Table4 . Since some macroblocks in the P and B frames do not have motion vectors and are intracoded independently, percentage of the motion compensated macroblocks is generally smaller than the percentage of P , B frames.... ..."

Cited by 1