### Table 6. Changes in pixel colors

"... In PAGE 14: ... In order to measure the accuracy of a tem- poral antialiasing technique, we measure how many pixels changed their color by a given value. Such a histogram analysis is shown in Table6 . Each entry in that table counts the number of pixels that change their value by the range speci ed in the left column, accumulating the pixels of one image column, over 400 con- secutive frames.... ..."

### Table 2: Compression results on new image test set (in bits/pixel averaged over color planes)

2000

"... In PAGE 26: ...PEG-LS due to the more elaborate data format (e.g., marker segments, bit stu ng, etc.). Table2 shows (lossless) compression results of LOCO-I, JPEG-LS, and LOCO-A, compared with other popular schemes. These include FELICS [23], for which the results were extracted from [16], and the two versions of the original lossless JPEG standard [19], i.... In PAGE 26: ... These results are extracted from [16]. The images in Table2 are the subset of 8 bits/pixel images from the benchmark set provided in the above Call for Contributions. This is a richer set with a wider variety of images, including compound documents, aerial photographs, scanned, and computer generated images.... In PAGE 26: ... Compression ratios are given in bits/pixel averaged over color planes (for consistency, PNG was run in a plane-by-plane fashion). The results in Table2 , as well as other comparisons presented in [1], show that LOCO-I/JPEG-... In PAGE 27: ... However, they do provide a good indication of relative practical complexity. Table2 also shows results for LOCO-A. The comparison with CALIC shows that the latter scheme maintains a slight advantage (1-2%) for \smooth quot; images, while LOCO-A shows a signi cant advantage for the classes of images it targets: sparse histograms (\aerial2 quot;) and computer-generated (\faxballs quot;).... In PAGE 29: ... For example, in a CMYK representation, the runs in the K plane tend to be longer than in the other planes, so it is best to encode the K plane in a di erent scan. The performance of JPEG-LS, run in line-interleaved mode on the images of Table2 , is very similar to that of the component-by-component mode shown in the table. We observed a maximum compression ratio deterioration of 1% on \gold quot; and \hotel, quot; and a maximum improvement of 1% on \compound1.... ..."

Cited by 95

### Table 1. The value of the first five eigenvalues, their relative contribution and their cumulative contribution to the total variation (sum of all eigenvalues) of the color constant spectra and of normalized color constant spectra. color constant spectra normalized spectra

"... In PAGE 5: ...he spectral images are analyzed in two ways. One is based on color constant spectral images as described in section 3.2, the other on normalized color constant spectral images as described in section 3.3 For the color constant spectra, the first three eigenvectors of the sum and squares and cross products matrix contained more than 99% of the total variation as can be seen by the cumulative contribution of the eigenvalues ( Table1 ). Therefore the 170 bands of every pixel can be reduced to 3 bands (principal components), while maintaining over 99% of the original variation.... ..."

### Table 1. The value of the first five eigenvalues, their relative contribution and their cumulative contribution to the total variation (sum of all eigenvalues) of the color constant spectra and of normalized color constant spectra. color constant spectra normalized spectra

"... In PAGE 5: ...he spectral images are analyzed in two ways. One is based on color constant spectral images as described in section 3.2, the other on normalized color constant spectral images as described in section 3.3 For the color constant spectra, the first three eigenvectors of the sum and squares and cross products matrix contained more than 99% of the total variation as can be seen by the cumulative contribution of the eigenvalues ( Table1 ). Therefore the 170 bands of every pixel can be reduced to 3 bands (principal components), while maintaining over 99% of the original variation.... ..."

### Table 1. The value of the first five eigenvalues, their relative contribution and their cumulative contribution to the total variation (sum of all eigenvalues) of the color constant spectra and of normalized color constant spectra. color constant spectra normalized spectra

"... In PAGE 5: ...he spectral images are analyzed in two ways. One is based on color constant spectral images as described in section 3.2, the other on normalized color constant spectral images as described in section 3.3 For the color constant spectra, the first three eigenvectors of the sum and squares and cross products matrix contained more than 99% of the total variation as can be seen by the cumulative contribution of the eigenvalues ( Table1 ). Therefore the 170 bands of every pixel can be reduced to 3 bands (principal components), while maintaining over 99% of the original variation.... ..."

### Table 4. Arithmetic operations required by the proposed algorithm to estimate two missing color components at (a) a red/blue sampling position or (b) a green sampling position

2006

"... In PAGE 11: ... It was found that the final demosaicing results were more or less the same as the one without corruption in terms of both CIELab color difference and CPSNR. Table4 shows the complexity of the proposed algorithm. Note that some intermediate computation results can be reused during demosaicing and this was taken into account when the complexity of the proposed algorithm was estimated.... In PAGE 11: ...verage CPSNR of 39.89 dB and an average CIELAB color difference of 1.6007 in our simulations. Its complexity is also shown in Table4 . In our simulations, the average execution times for the proposed algorithm and its simplified version to process an image on a 2.... In PAGE 14: ... Performance of various algorithms in terms of CIELAB color difference Table 3. MSE contributed by different groups of pixels Table4 . Arithmetic operations required by the proposed algorithm to estimate two missing color components at (a) a red/blue sampling position or (b) a green sampling position ... ..."

Cited by 4

### Table 3: Relative color differences and standard deviations for components of color difference averaged over the 99 color samples of Set II. L denotes lightness difference, H ab denotes the hue difference, C ab denotes the chroma difference. Results are given for default (NTSC) vs. calibrated color matching matrix.

"... In PAGE 4: ... Cali- brated Transformation Matrices for Set II For 99 images of set II, independent samples are taken in a 3x3 neighborhood, for which the mean E and standard deviation is computed. Table3 summa- rizes the relative CIELAB E error for different con- version matrices, individual CIELAB pixel values are compared to the average CIELAB value. The top row shows the mean Eab using the de- fault NTSC conversion matrix.... In PAGE 4: ...o 3.5 and 2.2, respectively, as is shown in the bottom row. We conclude from Table3 that The structure and texture of textile causes fluc- tuations in the intensity of the image. However, L , H and C ab are of equal order of magnitude ( H ab is not an angular measure).... In PAGE 4: ... The standard deviation is due to the texture of the material. Table 4 shows similar results as Table3 , this time the results are obtained by comparison of pixel CIELAB values with the absolute CIELAB values measuredby a spectophotometer. The spectophotome- ter values are considered as ground-truth.... ..."

### Table 3: Relative color differences and standard deviations for components of color difference averaged over the 99 color samples of Set II. L denotes lightness difference, H ab denotes the hue difference, C ab denotes the chroma difference. Results are given for default (NTSC) vs. calibrated color matching matrix.

"... In PAGE 4: ... Cali- brated Transformation Matrices for Set II For 99 images of set II, independent samples are taken in a 3x3 neighborhood, for which the mean E and standard deviation is computed. Table3 summa- rizes the relative CIELAB E error for different con- version matrices, individual CIELAB pixel values are compared to the average CIELAB value. The top row shows the mean Eab using the de- fault NTSC conversion matrix.... In PAGE 4: ...o 3.5 and 2.2, respectively, as is shown in the bottom row. We conclude from Table3 that The structure and texture of textile causes fluc- tuations in the intensity of the image. However, L , H and C ab are of equal order of magnitude ( H ab is not an angular measure).... In PAGE 4: ... The standard deviation is due to the texture of the material. Table 4 shows similar results as Table3 , this time the results are obtained by comparison of pixel CIELAB values with the absolute CIELAB values measuredby a spectophotometer. The spectophotome- ter values are considered as ground-truth.... ..."

### Table 3: Color code definitions for difference images in SR detection. Pixel color Indicates

"... In PAGE 6: ... A visual comparison of the results is facilitated by recording the number of SR pixels labeled by each approach and generating a color-coded difference image, examples of which are shown in Figure 4(e). The color codes are defined in Table3 . For almost all of the cases in the 120 cervigrams, both SR detection approaches were subjectively found to be equally accurate by the three NLM researchers.... ..."

### Table 8: Color code definitions for difference images in SR classification Pixel color Indicates

2006