### Table 2. Ground truth comparison of light estimation for specular bottle image

2003

"... In PAGE 7: ... The false detections in (d) and (h) do not exhibit much inter-cue or intra-cue consistency, so these two light directions are not strongly supported. Table 1 and Table2 list comparisons of the estimated directions and intensities with ground truth obtained from high dynamic range images of a mirrored sphere. The light direction is expressed as a unit vector in a coordinate system centered on the sphere.... ..."

Cited by 8

### Table 1: Change detection in image intensities vs. change detection in normal flow fields Change detection in intensities Change detection in normal flow

"... In PAGE 8: ...ajority of the points in a small neighborhood. This allows isolated points to be removed. There is a number of interesting analogies that can be drawn between change detection methods that are used to detect moving objects in the field of view of a static observer and the proposed method for the detection of changes in 3D motion. Table1 summarizes these analogies. 3.... ..."

### Table 1: Change detection in image intensities vs. change detection in normal flow fields Change detection in intensities Change detection in normal flow

"... In PAGE 8: ...ajority of the points in a small neighborhood. This allows isolated points to be removed. There is a number of interesting analogies that can be drawn between change detection methods that are used to detect moving objects in the field of view of a static observer and the proposed method for the detection of changes in 3D motion. Table1 summarizes these analogies. 3.... ..."

### Table 1: Detection results of experiment 1. a) is the results with color information, b) is the results with only grayscale information.

2001

"... In PAGE 14: ...The results of experiment 1 are summarized in Table1 -a). Each cell in a table shows the percentage of correct detection out of 20 test cases per person.... In PAGE 14: ... Each cell in a table shows the percentage of correct detection out of 20 test cases per person. The first row of the Table1 -a) shows 100% correct detection with randomly transformed neutral faces. The second row shows 100% correct detection with randomly transformed smiling faces, and the third row shows 96% correct detection with occluded faces.... In PAGE 14: ... a) is the results with color information, b) is the results with only grayscale information. Table1 -b shows the results of the same experiments with color information quot;turned off quot;. The results show slight degradation in the detection performances.... ..."

Cited by 3

### Table 1: Coefficients fhk;lg of the Markov random field wood model [14].

1997

"... In PAGE 21: ... The 64x64 pixels were divided into groups g1 and g2: g1 contains the pixels in the upper left and lower right of the image, and g2 contains the pixels in the diagonal band running through the center of the image. The prior model for g1 is the wood model of Table1 ; the prior model for g2 uses the same coefficients in Table 1, but with the table rotated by 90 degrees. The cross correlation between groups g1 and g2 is zero.... ..."

Cited by 34

### Table 1: Coefficients a177 a12a11a32a250a14a13 a31a138a179 of the Markov random field wood model [14].

"... In PAGE 21: ... The 64x64 pixels were divided into groups a198 a24 and a198a66a27 : a198 a24 contains the pixels in the upper left and lower right of the image, and a198a66a27 contains the pixels in the diagonal band running through the center of the image. The prior model for a198 a24 is the wood model of Table1 ; the prior model for a198a66a27 uses the same coefficients in Table 1, but with the table rotated by 90 degrees. The cross correlation between groups a198 a24 and a198 a27 is zero.... In PAGE 21: ... The 64x64 pixels were divided into groups a198 a24 and a198a66a27 : a198 a24 contains the pixels in the upper left and lower right of the image, and a198a66a27 contains the pixels in the diagonal band running through the center of the image. The prior model for a198 a24 is the wood model of Table 1; the prior model for a198a66a27 uses the same coefficients in Table1 , but with the table rotated by 90 degrees. The cross correlation between groups a198 a24 and a198 a27 is zero.... ..."

### Table 2: Gibbs Sampler timings for a binary (G = 2) image (execution time in seconds per iteration on a CM-5 with vector units) 3.2.1 Iterative Gaussian Markov Random Field Sampler The Iterative Gaussian Markov Random Field Sampler is similar to the Gibbs Sampler, but instead of the binomial distribution, as shown in step 3.2 of Algorithm 1, we use the continuous Gaussian Distribution as the probability function. For a neighborhood model N, the conditional probability function for a GMRF is:

in Scalable Data Parallel Algorithms for Texture Synthesis and Compression using Gibbs Random Fields

1995

"... In PAGE 14: ...omplexity, and steps 2.3 and 2.4 in O(G np ) computational time, yielding a computation complexity of Tcomp(n; p) = O( n (G+jNsj) p ) and communication complexity of 8 gt; lt; gt; : Tcomm(n; p) = O(jNsj qnp ), on the CM-2; Tcomm(n; p) = O(jNsj( (p) + qnp )), on the CM-5, per iteration for a problem size of n = I J. Table 1 shows the timings of a binary Gibbs sampler for model orders 1, 2, and 4, on the CM-2, and Table2 shows the corresponding timings for the CM-5. Table 3 presents the timings on the CM-2 for a Gibbs sampler with xed model order 4, but varies the number of gray levels, G.... ..."

Cited by 7

### Table 1: Number of sampling locations for various fea- ture types. MRF: Markov Random Field model, IG: in- tensity gradient. Gabor and IG are extracted only from femur images.

2004

"... In PAGE 3: ... Markov Random Field (MRF) texture model ex- tracts features from moderate-sized sampling regions. In the current implementation, the number of sampling lo- cations is set as shown in Table1 . Figure 2 illustrates an example of adaptive sampling at the femoral neck.... ..."

Cited by 4

### Table 1: Error rates in recognizing raw gray-scale images

2000

"... In PAGE 4: ...5) and this imposes a practical limit on the maximum number of gray levels usable for a given n-tuple size. Table1 shows recognition error rates achievable using 2, 4 and 8 gray levels per pixel. Both the n-tuple and the MWC classifier are configured with 500 3-tuples or 4-tuples.... ..."

Cited by 2

### Table 4 Comparison of Models with Weighting Based on PAYROLLa

2001

"... In PAGE 12: ... Thus, the logarithmic transformation did not remove the heteroscedasticity. Table4 shows the results of fitting several mod- els, using PAYROLL to weight subject-specific variances. Because the logarithmic transforma- tion did not capture all of the heteroscedasticity, we fit the models using both PP and LnPP as response variables.... In PAGE 12: ... Plots of standardized residu- als from these models versus PAYROLL, not dis- played here, indicate that this weighting captures the heteroscedasticity. Among the models with LnPP as the response, Table4 shows that the model with YEAR as an additional random effects component, but not fixed effects, provides the best fit to the data. This model can be written as LnPPit 5 a1i 1 a2iYEARt 1 b1 1 eit~PAYROLLit!1/2, where {eit} is an i.... In PAGE 12: ...here {eit} is an i.i.d. sequence of noise terms. Among the models with PP as the response vari- able, Table4 shows that the model with no addi- tional random or fixed components is the pre- ferred choice. This model is PPit 5 a1i 1 b1 1 eit~PAYROLLit!1/2.... ..."

Cited by 6