Information Technology Reference
In-Depth Information
level is visible more clearly in a longer camera base distance. Ratings of visual
strain and perceived depth increase only when increasing the camera base dis-
tance. However, if the crosstalk level increases, visual strain and perceived depth
might not change accordingly. Furthermore, Kim et al. [13] has proposed an ob-
jective metric by taking into account both acquisition and display issues. For
instance, using a multiple cameras structure may cause impairment such as mis-
alignment. The experimental results demonstrated that a depth map is a useful tool
to find out implied impairments, in which the depth map was obtained by estimat-
ing disparity information from stereoscopic videos. By using the depth map, the
depth range, vertical misalignment and temporal consistency can be modeled
separately to exhibit different viewing aspects. They are then integrated into a
metric by a linear regression, which can predict the levels of visual fatigue.
Existing work on perceptual quality evaluation for both video-plus-depth and
multi-view video 3D presentation is mostly focused on assessing the quality deg-
radation caused by compression errors. Currently, most 3D compression schemes
are developed for stereoscopic images or videos that consist of two views taken
from a lightly different perspective in a 3D scene. Since one image (target) in a
stereo-pair images can be restored from the disparity information and the other
one image (reference), the reference image is in general coded with a traditional
2D compression scheme whereas the target image can be represented by disparity
vectors. Stereoscopic coding schemes using the disparity estimation can be classi-
fied into: 1) intensity-based methods and 2) feature-based methods [1]. Although
many quality metrics for 2D image quality assessment have been proposed, the
quality models on stereoscopic images have not been widely studied. Hewage et
al. [14] tested the performance of three quality metrics, including peak signal-to-
noise ratio (PSNR), video quality model (VQM) proposed by NTIA [15], and
structural similarity model (SSIM) [16], with respect to a subjective quality ex-
periment on a series of coded stereoscopic images. The experimental results dem-
onstrated that VQM is better than other two metrics while its performance is still
not promising. Similar work has been done in [17]. Four metrics, as well as three
approaches, called average approach, main eye approach, and visual acuity ap-
proach, were tested for evaluating the perceptual quality of stereoscopic images.
Further, disparity information was integrated into two metrics for the quality as-
sessment [18]. It was found that the disparity information has a significant impact
on stereoscopic quality assessment, while its capability has not been studied ade-
quately. In [19], only absolute disparity was used. It was found that added noise
on the relatively large absolute disparity has greater influence than on other dis-
parity. Subsequently, a metric called stereo sense assessment (SSA) based on the
disparity distribution was proposed.
In addition, some special metrics that take into account advantage of the char-
acteristics of 3D images have been proposed. Boev et al. [20] combined two com-
ponents: a monoscopic quality component and a stereoscopic quality component,
for developing a stereo-video quality metric. A cyclopean image for monoscopic
quality, a perceptual disparity map, and a stereo-similarity map for stereoscopic
quality were defined. These maps were then measured using SSIM in different
scales and combined into a monoscopic quality index and a stereoscopic quality
