### Table2. A comparison of the recognition accuracy (%) of audio-only and bimodal recognizers in presence of noise. The audio-only recognizer uses MFCC-derived features. The Bimodal1 recognizer uses (G_norm + P_norm) feature computed based on ROI R1. The Bimodal2 recognizer uses the same visual feature computed based on R2 ROI that is compensated for scaling and translation.

2001

"... In PAGE 5: ... They were added to the audio channel systematically at various SNRs, and only in the testing data. Table2 summarizes the results. We can observe that the bimodal recognizers consistently outperformed the audio-only counterpart at all SNRs.... ..."

Cited by 5

### (Table IX). Comparing our results to reported p-norm results (Table X) shows

1991

Cited by 176

### Table 5.3 Global super-errors for p in the norm jjj jjj0

### Table 1 The PSNR comparison for the original (PSNR-OR), the standard hard thresholding(PSNR- HD), the TV hard thresholding (PSNR-TVHD), the TV log function (PSNR-TVLOG), and the TV p-norm (PSNR-TVP) images

"... In PAGE 16: ... We display the m = 64 64 nonzero coe cient reconstruction calcu- lated by the standard hard thresholding on the right of Fig 11, and the reconstructions of TV wavelet thresholding model by hard thresholding approximation (11) on the left of Fig 12 (with = 5), the log function approximation (15) (with = 3:2; = 0:02) on the right of Fig 12, and the p-norm approximation (16) (with = 3:2; = 0:02) on the left of Fig 13. We also show the standard PSNR measurements of the displayed images in Table1 . The PSNR is de ned by PSNR = 10 log10( 2552 ku u0k2 2 )(dB); where 255 is the maximum intensity value of gray scale images, u0 the noise free original image, and k k2 the standard L2 norm.... ..."

Cited by 1

### Table 1. Reconstruction errors [mm] using PSP and MRF regularization. The mean one std. is shown for each method. The site-prior is governed by the p-norm and q controls the sensitivity of the observational energy term dependent on the surface normals.

2003

"... In PAGE 7: ... A rank test shows the signi cance of the MRF regularization since a reduction in the coe cient is obtained for all subjects. The improvement in shape reconstruction is show in Table1 . Applying the observation model is performed with = 0:5.... ..."

Cited by 11

### Table 1. Comparison of delta-wye and edge-packing bounds for example graphs

"... In PAGE 20: ... The bounds for this case are 1 ? (1 ? p4)4 R(G4;4; p) [1 ? (1 ? p)4]4: These packing collections o er the best possible edge packing bounds for most reasonable probability assignments. The edge-packing bounds were computed and compared to the delta-wye bounds, using the p-norm, and the results are given in Table1 . The delta-wye bounds were 20 to 80 times tighter than the edge packing bounds for these three graphs.... In PAGE 22: ... It is evident, however that the delta-wye bounds again o er a dramatic im- provement over the lattice bounds, with p-norms 10{15 times smaller. Figure 7a{7c shows the plots of the reliability bounds for the examples given in Fig- ures 6a{6c, and the associated p-norm values are given in Table1 . Figure 7d shows an almost perfect t of the delta-wye best estimate to the exact values of the reliability for the graph GDO.... ..."

### Table 2 illustrates the similarity measures obtained from the new evidence counting method and the Euclidean distance measuring method, with the same query images. (R norm , P norm and L n and the dataset used are described in the next section). In calculating the Euclidean distance, an asymmetric simple match- ing method, which was identified as the most effective Euclidean distance measure criteria in ARTISAN experiments [4], was used.

1998

"... In PAGE 10: ... Table2 : Comparison of retrieval performance between the usage of evidence count criteria and Euclidean distance measure criteria. 4 Preliminary Results We have conducted preliminary experiments on the performance of the system using a smaller image database of 210 trademark images which includes nine groups of perceptually similar images (in total 61, some examples are shown in figures 6-11) which have been pointed out by trademark examiners during eval- uation experiments of the ARTISAN system [4], and 149 arbitrary selected images.... ..."

Cited by 5