Information Technology Reference
In-Depth Information
these IQMs in evaluating the stereoscopic image quality is similar to that in predict-
ing the 2D image quality. A better IQM on the 2D image quality assessment usu-
ally has better performance on the stereoscopic quality assessment. However,
according to the averages of different IQMs, as shown in the last rows in the Ta-
bles, the robustness of these IQMs to the change of distortion types in stereoscopic
image quality assessment is much worse than that in 2D image quality assessment.
The performance of these IQMs on the entire distortion types in the 2D image qual-
ity assessment is much better than that in the stereoscopic image quality assess-
ment. In our opinion, the reason is that the perceived quality is not only affected by
image content, but other attributes of stereopsis, such as disparity, have significant
influence on the quality evaluation of the stereoscopic images as well.
3.3 Perceptual Stereoscopic Quality Assessment Based on
Disparity Information
Human eyes are horizontally separated by about 50-75 mm (interpupillary dis-
tance) depending on each individual. Thus, each eye has a slightly different view
of the world. This can be easily seen when alternately closing one eye while look-
ing at a vertical edge. At any given moment, the lines of sight of the two eyes
meet at a point in space. This point in space projects to the same location (i.e. the
center) on the retinae of the two eyes. Because of different viewpoints observed by
the left and right eyes however, many other points in space do not fall on corre-
sponding retinal locations. Visual binocular disparity is defined as the difference
between the points of projection in the two eyes and is usually expressed in de-
grees as the visual angle. The brain uses binocular disparity to extract depth
information from the two-dimensional retinal images in stereopsis. In computer
stereo vision, binocular disparity refers to the same difference captured by two dif-
ferent cameras instead of eyes [43]. Generally, one image of stereo-pair images
can be restored from the disparity and the other one image. Therefore, we believe
that the disparity between the left and right eye images has an important impact on
visual quality assessment. In this subsection, we apply the disparity information in
the stereoscopic image quality assessment [44].
In this work, we do not intend to study the estimation methods of disparity map
between a stereo-pair images and their impact on the quality assessment. We
chose a belief propagation based method to estimate the disparity map [45]. Be-
cause of the distorted regions, the disparity of the original stereo-pair images is
different from that of the distorted stereo-pair images, even though the relative po-
sitions of the objects in the image pair do not change at all. Figure 8 shows exam-
ples of an original
Art
image (right eye), distorted images, and the corresponding
disparity maps. Because the real objects in the image do not change during the dis-
tortion process, changes between two disparity images (one is original disparity
and another is the disparity between the left eye image and the distorted right eye
image) are usually located at those positions where the distortions are clearly visi-
ble, such as noise added regions, regions with blockiness. Consequently, we can
compare the disparity images to obtain a quality prediction for the distorted stereo-
scopic images.
