### Table 2. Mean and Standard Deviation for Wavelet-Based Features of Veg1 and Veg2 for Noisy and Denoised Temporal Signatures

"... In PAGE 9: ... Veg1 increases from 190,165 to 251,508, representing a change of 32%, and for class Veg2 increases from 102,728 to 118,408, representing a change of 15%. From Table2 , one can see that similar types of changes occurred for the other wavelet- based features. That is, the denoising process causes class distributions to change sig- nificantly in the wavelet-based feature space, and as a result, one would expect the classification accuracies to also change, depending on the type of classifier used.... ..."

### Table 2: List of wavelets bases used in several turbulence studies.

"... In PAGE 7: ...he experimental setup can be found in (see Szilagyi et al., 1996). 4 Results and Discussion As discussed in Farge (1992), the wavelet basis function can inject properties that are a function of the wavelet choice rather than the process under consid- eration. Currently, the choice of the wavelet basis function in the analysis of turbulence measurements is arbitrary (see Table2 for summary of some studies). This motivated us to consider two elements in the wavelet thresholding process: the choice of the criterion and the choice of the wavelet basis.... ..."

### Table 4. Classification Accuracies for Wavelet-Based Features, in percent

"... In PAGE 9: ... Table 3 provides the classification results for the Fourier-based features when using the NM and NN classifiers, for both the noisy and denoised temporal signa- tures. Likewise, Table4 provides the classification results for the wavelet-based fea- tures when using the NM and NN classifiers, for both the noisy and denoised temporal signatures. From Table 3, one can see that the classification results of the noisy temporal signatures were unchanged by applying the denoising process, remaining 81% and 67% for the NM and NN classifiers, respectively.... In PAGE 10: ...179 surprising, considering the fact that the denoising process did not cause the class distributions to significantly change in the Fourier-based feature space. However, from Table4 , it is apparent that the denoising process improved the classification results for the wavelet-based features, increasing the overall classification accuracies from 95% to 100% for both the NM and NN classifiers. Also, by comparing Tables 3 and 4, it is apparent that the wavelet-based features significantly outperform the Fou- rier-based features.... ..."

### Table 2: Errors of the new wavelet-based histogram and that of the old wavelet-based histogram for query Type A using

"... In PAGE 8: ... We present results for Type A queries using the synthetic data described in #5B16#5D. Table2 shows various types of errors for four data sets. In the experiments, all data sets are generated using Zipf distribution with di#0Berent Zipf pa- rameters.... ..."

### Table 3. Objective fusion performance of Fig4(a) and Figure4(b) PNN RBF Wavelet-based

"... In PAGE 5: ... The decomposed block is with size of 32 32. The objective evaluations of the results are shown in Table3 . The objective evaluation of DWT-based method is also given.... ..."

### Table 3. Results using the wavelet based approach on a synthetic dataset. Shows the running time with/without the optimization.

"... In PAGE 11: ... We then proceeded to test the wavelet based method for optimizing the mining speed (Section 4). Table3 presents our results. The results demonstrate that the wavelet based method significantly improves the mining speed if there are clusters through which frequent trajectories of the desired length m pass, since it provides a quick method of finding which trajectories should be eliminated from further processing.... ..."

### Table 4 t-statistics for comparing mean performances of two groups (the results using the wavelet-based indicators versus the results using the technical indicators for RC30/[35%, 50%])

"... In PAGE 11: ...nd 0.6574, respectively, obtained in the earlier study (see the last row in Table 5). A statistical two-tailed unpaired t-test has been applied to determine whether the results differences between two groups are statistically significant. Shown in Table4 are t-values and their corresponding p-values under each measure, indicating that the generated GDTs statistically exhibit much better performance under all of three criteria at a significant level of aB/1.0C29/10C287.... ..."

### Table 2: Wavelet-based multiresolution analysis

Cited by 2

### Table 5: Comparison with wavelet-based methods

"... In PAGE 21: ...he feature extractions can be seen in Subsection 4.2. In order to show more comparability, all the methods use CLDA and the NN classifier with the Euclidean distance. Table5 shows that the LDC based feature extraction has the best result on the ORL database, both data sets of the FERET database, and it outperforms WaveletFace, though it underperforms LDB and MLDB marginally on the CMU-lights database. Moreover, LDC is more efficient than LDB, MLDB because of the lower computational complexity when K is large, especially on the large FERET data set (K = 255).... In PAGE 22: ... The results on the ORL database, both data sets of the FERET database and the CMU-lights database are shown in Table 6. Comparing the results on Table 6 with Table5 which uses the Euclidean distance, it shows that the triangle square ratio criterion performs better than the Euclidean distance considerably on both data sets of the FERET database and the CMU-lights database, while its efficacy is very close to the Euclidean distance on the ORL database. In fact, the FERET database, the CMU-lights database concern about illumination variations (light intensity and direction respectively), and the ORL database concerns about expression and pose changes.... ..."