### Table 2: Hausdorff + Neural Network (0.85)

"... In PAGE 4: ... Such symbols were not present in the train- ing map. Table2 shows the recognition results when trained neural networks are also used, with a measure of confi- dence of 0.85.... ..."

### Table 3: Hausdorff + Neural Network (0.50)

"... In PAGE 4: ... For this symbol, more training would thus be necessary. Table3 shows the result when using trained neural networks with a measure of confidence of 0.5, where can- didates will be accepted when there is more confidence that they are symbols than false positives.... ..."

### Table 4: Hausdorff + Neural Network (0.15)

"... In PAGE 4: ... Here only a few symbols are missed, and only a few false positives remain. Table4 shows the result when using trained neural networks with a measure of confidence of 0.15, where candidates are accepted when there is no compelling evi- dence that the symbol should be rejected.... ..."

### Table 1. Spearman correlations (CS) of subjective results with several metrics: Our structural distortion measure (MSDM), Hausdorff maximum, mean and root mean square (RMS) distances.

"... In PAGE 9: ...ean distance does not reflect at all this subjective opinion (Hausdorff mean=0.16 vs 0.25). Table1 presents the performances of our measure and Hausdorff metrics in term of Spearman correlation CS with subjective results. Overall the proposed MSDM outperforms Hausdorff distances, particularly for Venus and Dyno objects.... ..."

### Table 1: Measurements for the used models. From left to right: number of vertices/edges in original and base model, construction time for simplification with planarity measure, primal quadric measure, Hausdorff-measure with maximum of six samples per assignment list and if available not restricted Hausdorff-measure. The first five models are polygonal and the other five triangular.

2004

"... In PAGE 16: ... More examples are shown in Figure 11. Table1 tabulates the results of further measurements. The first five models are polygonal, whereas the other five models are purely triangular.... ..."

### Table 6. The estimated number of operations required for the Hamming and Hausdorff distance based key VOP selection algorithms. Hamming distance measure does not include the operations involved in computing the activity level.

2000

"... In PAGE 13: ... If we approximate the number of contour blocks by 2x(N+M), then this algorithm would require 16x(N+M)x(N+M) square, 8x(N+M)x(N+M) square-root, and 8x(N+M)x(N+M) addition operations for each VOP candidate. Table6 summarizes the estimated number of operations required for each algorithm and shows an example for typical N and M values. Even when the number of operations required to estimate the threshold for the Hamming distance based algorithm is considered, the Hausdorff distance based algorithm requires significantly more operations than that of the Hamming distance based algorithm.... In PAGE 22: ...able 5. Temporal segments for the Weather, Bream, and Hall Monitor video objects. .......................... 24 Table6 . The estimated number of operations required for the Hamming and Hausdorff distance based key VOP selection algorithms.... ..."

Cited by 16

### Table 1: Statistics on some remeshed models. Con- trast indicates the impact of curvature on sizing. Ap- proximation error measures the Hausdorff distance between the resulting mesh and the input as a percent of the bounding box diagonal.

2003

"... In PAGE 11: ... (a) (b) (c) (d) Figure 8: The effect of segmentation. The statistics of the results appear in Table1 . (a) Input tree mesh.... ..."

Cited by 8

### Table 1: Novelty measures based on language models.

2003

Cited by 22