### Table 8: Improved coverage estimates for combinational benchmarks. exact cov. (%) Stafan (short) Stafan (long) new algorithm (both sequences)

Cited by 1

### Table 1. Comparison of results between grids with and without diagonals. New results

1994

"... In PAGE 2: ... For two-dimensional n n meshes without diagonals 1-1 problems have been studied for more than twenty years. The so far fastest solutions for 1-1 problems and for h-h problems with small h 9 are summarized in Table1 . In that table we also present our new results on grids with diagonals and compare them with those for grids without diagonals.... ..."

Cited by 11

### Table 1: New Matrix Pencil Algorithm

1998

"... In PAGE 10: ... It is worth mentioning that our matrix pencil algorithm has the same order of complexity as the KT algorithm [3] and Hua-Sarkar apos;s matrix pencil algorithm [6]. As a matter of fact extra computations in our matrix pencil algorithm come from the Hankel approximation part (Step 2 in Table1 ), which requires several SVD apos;s until the algorithm converges. But for most practical cases of interest the Hankel approximation typically converges within a few iterations.... In PAGE 11: ... Also Figure 1 (c) shows that when estimating !2 the noise threshold of the new matrix pencil algorithm is about 10 dB lower than that of the KT algorithm and is about 5 dB lower than those of MKT and Hua-Sarkar apos;s matrix pencil algorithms. It was mentioned before that to reduce the noise e ect, in our algorithm we have preprocessed the noisy data to minimize the noise e ect (Step 2 in Table1 ) by performing rank de cient Hankel approximation. In order to see the e ect of this preprocessing on the performance of the original Hua-Sarkar matrix pencil algorithm, we also employed the preprocessed data in the Hua-Sarkar matrix pencil method.... ..."

Cited by 4

### Table 2: The average misclassification rates with the new LDB algorithm using the simple histgrams as the em- pirical pdf estimation method (averaged over 10 simula- tions). Here, NLDB5xxx means that m = k = 5 in (11).

### Table 2: The average misclassification rates with the new LDB algorithm using the simple histgrams as the em- pirical pdf estimation method (averaged over 10 simula- tions). Here, NLDB5xxx means that a143a21a25

in Key Words: Local Feature Extraction, Pattern Classification, Projection Pursuit, Density Estimation

### Table 2: The true and estimated poles of the transfer function H(z)

1998

"... In PAGE 13: ... Figures 3 (b), (c) and (d) show the results. Table2 shows the mean and variance of the estimated poles of the transfer function using KT, MKT and the new matrix pencil algorithms. From Figure 3 and Table 2, it is clear that the new matrix pencil algorithm outperforms the KT and MKT algorithms.... In PAGE 13: ...nd the new matrix pencil algorithms respectively. Figures 3 (b), (c) and (d) show the results. Table 2 shows the mean and variance of the estimated poles of the transfer function using KT, MKT and the new matrix pencil algorithms. From Figure 3 and Table2 , it is clear that the new matrix pencil algorithm outperforms the KT and MKT algorithms. 5 Conclusion In this paper a new matrix pencil algorithm for estimating the parameters (frequencies and damping factors) of exponentially damped sinusoids in noise is proposed.... ..."

Cited by 4

### Table 3.2: Summary of experimental results: Optimal is numerically estimated. GFI is the performance of the new algorithm after 200 iterations using automatic range expansion when it produces improved results (automatic range expansion made no signi cant improvements for ranges less than 10000). RFI is from traditional table lookup with random trials. The last column indicates how many trials RFI needs to equal GFI apos;s performance.

### Table 3: The parameter values for segmentation of linescan images. 4 SUMMARY We have formulated a general image processing system for detection and segmentation of product inspection items in X-ray imagery. The detection and segmentation steps handle items at any orientation and are fast. A new technique was developed to estimate the number of items in a cluster and the center of each. These were used in a new watershed algorithm that is robust to gray-scale variations in objects, that can handle a large degree of overlap between touching objects, and avoids oversegmentation errors associated with other techniques. The general algorithm was developed on one database and extended with minor di erences to 3 new databases with excellent results. Larger versions of the last 3 databases are needed to nalize the algorithm for them.

"... In PAGE 10: ...3 X-Ray Linescan Images: Parameter Selection and Results As mentioned earlier, the sizes of the items in the linescan images for the 3 new databases are around 70% of those in the X-ray lm images. Hence, the 5 and peaksort window segmentation parameters ( Table3 ) were reduced to around 70% of the corresponding values for the X-ray lm images. The B and D values di er, but the same T = (B + D)=2 rule was used (Table 3).... In PAGE 10: ... Hence, the 5 and peaksort window segmentation parameters (Table 3) were reduced to around 70% of the corresponding values for the X-ray lm images. The B and D values di er, but the same T = (B + D)=2 rule was used ( Table3 ). The peaksorting thresholds TPS were selected as before, based on the lowest peak on a nut in each database (see Table 3).... In PAGE 10: ... The B and D values di er, but the same T = (B + D)=2 rule was used (Table 3). The peaksorting thresholds TPS were selected as before, based on the lowest peak on a nut in each database (see Table3 ). The minimum true peak values for X-ray linescan images are di erent from those in X-ray lm images since the linescan images tend to be darker with a higher degree of noise than in X-ray lm images.... ..."