### Table 1: Correspondence of the isometry operator in the spatial and wavelet domains.

1997

Cited by 3

### Table 5: An experimental comparison of direct vs. wavelet convolution. The DC columns report direct convolution times, the WC stands for wavelet-domain convolution, the SC stands for wavelet-domain convolution using symmetry. All times are in milliseconds for a single color channel. The WC and SC times include the final reconstruction step, necessary for displaying the convolved result.

Cited by 3

### Table 5: An experimental comparison of direct vs. wavelet convolution. The DC columns report direct convolution times, the WC stands for wavelet-domain convolution, the SC stands for wavelet-domain convolution using symmetry. All times are in milliseconds for a single color channel. The WC and SC times include the final reconstruction step, necessary for displaying the convolved result.

2001

### Table 4: Results for initial fractal block coding in the wavelet domain. Level 2 to Level 1, domain block size 4, 4-block radius search, basis projection not enabled.

"... In PAGE 31: ... Level 2 to Level 1, domain block size 4, no search, basis projection enabled. Table4 the e ects of enabling a 4 block radius search and gure 15 shows the rate distortion curve. Table 5 shows the e ect of the coe cient rate value on a system where the wavelet quantisation rate is held constant ( at 32 in this case), no search and the basis projection enabled.... ..."

### Table 5: Results for initial fractal block coding in the wavelet domain. Level 2 to Level 1, domain block size 4, no search, basis projection enabled.

"... In PAGE 31: ... Table 4 the e ects of enabling a 4 block radius search and gure 15 shows the rate distortion curve. Table5 shows the e ect of the coe cient rate value on a system where the wavelet quantisation rate is held constant ( at 32 in this case), no search and the basis projection enabled. Figure 17 shows the rate/distortion curve for the basic coder with no basis projection, with and without searching enabled.... ..."

### Table 2: Results for initial fractal block coding in the wavelet domain. Level 2 to Level 1, domain block size 4, no search, basis projection not enabled. Wavelet Rate Value MSE PSNR Bits per Pixel

"... In PAGE 30: ... 5.1 Results Table2 shows the results of using the coder ( with rst range coe cient coder ) with a rate value of 512 and varying the wavelet quantiser rate. Basis projection is not enabled and the coder does no searching for the best range block.... ..."

### Table 3: Test results for Expression Classification in terms of average false negative rate (FNR), false positive rate (FPR) and the standard deviations (STD) over 20 iterations. Average False Negative Rates (FNR) Spatial domain (step 1) Wavelet domain (step 2)

2006

"... In PAGE 14: ... 7. Experimental Results In Table3 , we display all the results that have been produced following the two experimental steps (Section 6.4) on the normalized face dataset described in Section 2.... In PAGE 18: ...1. Error rate improvement in relation to the wavelet subspaces My experimental results in Table3 show that D from the D-faces (D, Dx and Dy) of Wavelet AymmetryFaces produces the greatest improvements (86.2% on FNR and 93.... In PAGE 22: ... In this work, we have successfully investigated the implications of wavelet transforms on AsymmetryFaces. We have demonstrated that (1) by applying wavelet transforms on D-faces, a significant improvement can be achieved ( Table3 ); (2) certain subspaces of a wavelet tree have even more discriminative features compared to others, for instance higher frequency band (LH and HL from Table 4) of the wavelet tree; (3) the way S-faces are constructed, their image-intensity domain is already the optimal space with maximum discriminative features. Wavelet transforms are definitely useful at extracting features that can be used to improve classification rates.... ..."

### Table 1. Averages and standard deviations for the eyeglasses experiments using fusion in the wavelet domain. The columns represent the gallery set and the rows represent the test set. The first entry in each cell shows the average performance and standard deviation from the visible images, the second entry is from the IR images, and the third entry is from the fused images. The bottom entry shows the minimum and maximum recognition performances from the three cross-validation runs achieved when using the fused images. Test scenarios for which the test and the gallery sets had common subsets were not performed.

"... In PAGE 8: ... EXPERIMENTAL RESULTS Our experimental results illustrate clearly that IR is robust to illumination changes but performs poorly when glasses are present in the gallery set but not in the test set and vice versa. Considerable improvements in recognition performance were achieved in this case by fusing IR with visible images both in the wavelet (see Table1 ) and eigenspace (see Table 2) domains. The improvements were even greater when, in addition to eyeglasses, the test and the gallery set contained images taken under different illuminations.... ..."

### Tables 2 and 3 give the ideal thresholds for soft and hard shrinkage for the suite of synthetic signals studied by Donoho and Johnstone.9,8 These signals { called blocks, bumps, heavisine, and doppler { exhibit di erent types of spatially inhomogeneous behaviour. Table 4 gives the L2 risks achieved by using the ideal thresholds for the blocks signal. From the tables, it is evident that the ideal thresholds for soft shrinkage are much smaller than the ideal thresholds for hard shrinkage. The reason for this can be traced back to the nature of the risk function for soft shrinkage: large wavelet coe cients incur substantial risk due to bias. Ideal hard shrinkage typically has smaller risk than ideal soft shrinkage.5 Soft shrinkage su ers the most when the signal has very prominent local features which show up in the wavelet domain as large detail coe cients.

1995

Cited by 3

### Table 1: Comparison of Average PSNR (dB)

2002

"... In PAGE 3: ...The experimental results shown in Table1 and Figs. 4 and 5 in- dicate that our proposed RWTM method, which combines the ad- vantages of the wavelet-domain with irregular-mesh ME/MC, out- performs other ME/MC techniques operating in both the spatial and wavelet domains.... In PAGE 3: ... 4 and 5 in- dicate that our proposed RWTM method, which combines the ad- vantages of the wavelet-domain with irregular-mesh ME/MC, out- performs other ME/MC techniques operating in both the spatial and wavelet domains. In terms of average PSNR performance ( Table1 ), RWTM outperforms its nearest competitor (block-based ME/MC in the RDWT domain [3,11]) by 0.4 dB for both the fast- motion Football and the slow-moving Susie sequences.... ..."

Cited by 7