### Table I: Measures of feature strength used for feature detection with automatic scale selection.

### Table1. Mean angular errors and corresponding standard deviations of approximated optic ow using zeroth order normal ow. The results are computed at di erent sets of spatiotemporal ( ; ) and integration scales $. The last two rows show results produced with the automatic scale selection method discussed in Sec. 5.

2001

"... In PAGE 14: ... We have computed the optic ow using xed scales and automatic scale se- lection in order to see how sensitive the algorithm is to scale changes. The results can be found in Table1 and from these we can see that in general automatic scale selection improves the results and from the small standard deviations we can conclude that scale selection removes outliers, but for some properly chosen scales the xed scale results are in some cases better. Furthermore, for certain xed ne scales we nd that the rst order model does not produce better results than the zeroth order model.... In PAGE 14: ... Furthermore, the zeroth order optic ow does not handle the diverging trees sequence well, which is not surprising considering that this sequence consists of a type of motion which is poorly modelled by a zero order model. In Table 2 we show the best results from Table1 together with the best results of other optic ow techniques. We clearly see that both zeroth and rst order optic ow outperforms other methods for both test image sequences, except for the diverging trees sequence where the results of the zeroth order scale selected optic ow is mediocre, and the reason for this is again that the zero order model is a poor model of this type of ow.... ..."

Cited by 2

### Table 5 Performance comparison between different interest point selection methods (automatic scale) using the GWR network

2007

"... In PAGE 6: ... The acquired model of normality was then used to filter out abnormal visual features in images acquired during inspection of the arena containing either of two novel objects (again, the orange football or the grey box). Table5... In PAGE 7: ...01). The results in Table5 show that performance of both interest point detectors was worse than that obtained with a fixed scale for the case of the grey box ( clear agreement between novelty filter response and actual novelty status), but was kept at the same level for the orange ball (compare with Table 3). We surmise that this is because the grey box stands out less well from the grey floor than the orange ball, and furthermore has smaller details that attract interest points.... ..."

### Table 1. This table shows the measure of feature strength used by Lindeberg for feature detection with automatic scale selection using his -normalisation. We have calculated the corresponding values of H and DH using our de nition of . This table is a reproduction of a table from [17] with an extension of the columns for H, DH and topological dimension T of the features. Note that a simple relation exists between the fractal dimension DH and the topological dimension T. The relation between and T is not as straight forward. Feature type Normalized strength measure H DH T

2000

"... In PAGE 6: ... This suggests a simple relation between the topological dimension of the feature and a suitable choice of fractal dimension. In Table1 we have listed Lindeberg apos;s [17] suggested normal- ized measures of feature strength. For each feature we have calculated the H and DH values that correspond to his suggested values.... In PAGE 10: ... We have found a normalisation expression that has the Hausdor dimension as a parameter. Through this expression we have also found a relation between the topological dimension and the fractal dimension of the local image round a feature (see Table1 ). We conjecture (for future experimental testing): The topological dimension of the feature uniquely determines the scale- space normalisation parameter.... ..."

Cited by 8

### Table 1: Numerical values of some characteristic entities obtained at the central point of the image in Figure 3 using di erent amounts of additive Gaussian noise and automatic scale selection. Note the stability of the selected integration scale (proportional to sdet L) with respect to variations in the noise level , and that the selected local scale tQ increases with . Observe also the increasing di erence between the estimates of the normalized anisotropy ~ Q computed at the selected local scale, and at zero local scale (true value 0.600). The last two columns show the error in surface orientation n computed by monocular shape-from-texture under a speci c assumption about the surface texture (weak isotropy).

1996

"... In PAGE 15: ...3 it is shown that under a certain assumption about the surface texture (weak isotropy), the estimate of surface orientation is directly related to the normalized anisotropy ~ Q, and to the eigenvector of L corresponding to the maximum eigenvalue. Table1 illustrates the accuracy in estimates of ~ Q and surface orientation computed in this 4In these curves there is also a minimum in the signature of ~ Q at coarse scales. The reason why this occurs is that the higher-frequency sine component is suppressed much faster than the lower-frequency sine component.... In PAGE 29: ... The middle column shows the same cylindrical surface image that was used in the rst row in Figure 4. Here, 25% white Gaussian noise has been added; a noise level high enough to ensure that direct computations on unsmoothed data are bound to fail (compare with Table1 ). It is quite obvious that the adaptive multi-scale blob detection technique is able to handle this noise level without much di culty.... ..."

Cited by 44

### Table 8.1: Various scale sizes with full color and grayscale output. 2Actually, jpegvid automatically selects a depth of 8 when grayscale output is wanted.

### TABLE I PARAMETERS OF THE AUTOMATIC SELFBALANCING SCALE

### Table 1. Summary of image quality comparison results on the five scales (the automatically filtered image is substantially worse [-2], modestly worse [-1], approximately the same [0], modestly better [1], substantially better [2] than the manually filtered image) and corresponding statistical significances. Manual vs. Automatic Image Quality

"... In PAGE 9: ... The significance of differences in a paired comparison was determined by performing the paired signed-rank Wilcoxon test (22). RESULTS In the image quality comparison ( Table1 ), both readers rated the automatic filtering and the manual filtering similarly with no significant statistical difference. Reader 1 gave slightly better credit for automatic image selection than reader 2, due to stronger preference for bright foreground detail in spite of additional background noise.... In PAGE 16: ...16 CAPTIONS Table1 . Summary of image quality comparison results on the five scales and corresponding statistical significances (-2=substantially worse, -1=modestly worse, 0=approximately the same, 1=modestly better, 2=substantially better).... ..."

Cited by 1

### Table 1: Automatic Parameter Selections and Results

"... In PAGE 2: ... In the program, ASA functions as a wrapper around DDS/SIM (see Fig 1). Table1 below shows three run results over a 100 and 200 symbol long pair of DNA sequences within the parameter ranges: w = 3 to 14, d1 = 200 to 219, d2 = 20 to 23, d3 = 5 to 7, ci = -10, f = 3 to 17, m = 2, u = -5, p = -10, and r = -2. Table 1: Automatic Parameter Selections and Results ... ..."