Image Processing Reference
In-Depth Information
1 Introduction
The atractive 3D video applications such as 3D television (3DTV) and free-view point-video
(FVV) have led to numerous researches in 3D video display technologies. Using depth in-
formation of the scene, user can experience 3D perception on 3DTV and FVV enables users to
choose the desired scene view point interactively. Despite of rapid advances in 3D technology,
the quality evaluation of 3D contents without a full subjective test is still difficult and human
viewers are needed to judge the quality of images or videos that is a costly and time consum-
ing task. Objective quality assessment tools like peak signal-to-noise ratio (PSNR) and struc-
tural similarity index measure (SSIM) can evaluate the 2D image quality much faster without
human interference and can be implemented in a machine. Therefore, a reliable objective qual-
ity assessment tool for 3D applications is desirable. However, besides visual quality that rep-
resents the image quality regardless of depth information, several other aspects like depth
quality, naturalness, visual fatigue, and discomfort should be taken into account for 3D qual-
ity evaluation [ 1 ]. As these 3D aspects are still under investigation [ 2 , 3 ] and in the absence of
a reliable 3D quality metric, diferent studies atempted to evaluate the 3D quality by consid-
ering depth map properties [ 4 , 5 ].
Bosc et al. [ 6 ] examined the reliability of 2D image metrics in 3D evaluation considering the
artifacts of stereoscopic images generated from seven different depth image-based rendering
(DIBR) algorithms. Considering the results, they proposed objective measurements based on
analysis of shift of the contours and mean SSIM score. To investigate the correlation between
depth map quality as a grayscale image and 3D video quality, different probable artifacts are
applied to depth maps in Ref. [ 7 ] . In other similar research [ 8 ] , the performance of three qual-
ity metrics including PSNR, SSIM, and video quality metric on coded stereoscopic images is
compared with subjective test.
In this research, subjective quality assessment is utilized to measure the effect of artifacts
generated from different depth map upsampling approaches on the final reconstructed ste-
reoscopic image. Image upsampling is the method of increasing spatial resolution of images
and depth map upsampling is of great importance in 3D applications. The high-speed time-of-
light cameras extract reliable depth maps. However, the spatial resolution of depth maps is
relatively low in comparison with original images. Therefore, diverse depth map upsampling
approaches are provided to obtain high-resolution depth maps. Also, it is important to evalu-
ate the upsampling quality in order to realize upsampling performance on 3D content quality.
In this article, test depth maps are upsampled using seven well-known upsampling al-
gorithms as each method can yield different artifacts. Then, the quality of each upsampled
depth map is evaluated using different objective image quality assessment (IQA) tools. Also,
the subjective quality assessment is used to evaluate the effect of depth map upsampling ar-
tifacts on the reconstructed stereoscopic image quality. Investigating the similarity between
2D quality evaluation and 3D perception, we will search for the most accurate IQA tool(s) for
3D quality evaluation. Using the proper automatic objective IQA tool will help to predict the
quality of 3D image without using the expensive subjective test and even free of watching the
stereoscopic image.
Since it is difficult to investigate all methods, seven approaches are selected to be utilized in
this work. The bilinear upsampling uses average weighted of four neighboring pixels for inter-
polation to achieve upsampled depth map. A similar method called bicubic upsampling (BCU)
is based on 16 neighboring pixels. The bilateral upsampling (BU) [ 9 ] is a prevalent approach that
combines a spatial filter and a range filter to preserve the edge regions in upsampling process.
Another upsampling method based on the BU is joint bilateral upsampling (JBU) [ 10 ] which util-
Search WWH ::

Custom Search