A study on the relationship between depth map quality and stereoscopic image quality using upsampled depth maps - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 127

4 Experimental results
The quality performance of the 7 upsampling methods is evaluated using 10 test depth maps
from Middlebury stereo dataset [ 21 ] . The test RGB images and related depth maps are shown
in Figure 2 . In order to obtain low-resolution depth maps, the original data are downsampled
to lower resolution first and then the high-resolution depth maps are constructed using seven
upsampling methods ( Figure 3 ) .

FIGURE 2 (a) Test RGB and (b) depth maps provided by Middlebury.

FIGURE 3 Upsampled depth maps of aloe (a) and bowling (b) obtained by BLU, BCU, BU,
DTBU, JBF, ABU, and VBU in the scan order.

Table 1 represents the average subjective quality score of each upsampling method and
Table 2 shows the mean scores of different objective quality metrics on upsampled depth
maps. The results are derived from averaging the quality scores of the collection of 10 images.
As upsampling artifacts are important in edge locations, edge PSNR (E-PSNR) and nonedge
PSNR (NE-PSNR) are also computed.

Table 1
Average Subjective Measurement Data of Upsampled Depth Maps

BLU BCU BU JBU VBU ABU DTBU

MOS 3.76 3.64 3.89 3.84 4.03 3.46 3.99