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 123

izes both a color data and its low-resolution depth map. The variance - based upsampling (VBU)
[ 11 ] avoids the usage of the constant variance by computing a variance for each pixel block.
The disadvantage of the JBU is that it is sensitive to homogeneous regions and the weighting
function can be assigned a wrong variance in nonedge regions. To solve this problem, an adapt-
ive bilateral upsampling method (ABU) [ 12 ] has been proposed, where a large weight is assigned
to color image at edge pixels and a large weight is assigned to depth data at nonedge pixels.
To overcome the limitation in reducing blur at low-gradient edge regions in prior methods, a
distance transform - based bilateral upsampling (DTBU) [ 13 ] has been proposed.
The Pearson, Spearman, and Kendall correlations are three proper approaches for similarity
measurement between each objective assessment result and the subjective IQA. Comparing
the objective and subjective scores, it will be inferred that which objective IQA tools show
the most correspondence with human judgment and have superiority for 3D quality assess-
ment. DIBR or 2D + Depth is used to generate a stereoscopic image. Figure 1 shows the overall
framework that examines the relation between upsampling methods and 3D perception.
FIGURE 1 The flow diagram of experiment.
This article is organized as follows. In Section 2 , different IQA tools considered in this work
are described. Section 3 presents the subjective quality assessment methodology and the ex-
perimental results are provided in Section 4 . Finally, we summarize our work in Section 5 .
2 Objective quality assessment tools
Full-reference image quality assessment (FR IQA) compares test and reference images, there-
fore, both ground-truth and upsampled depth map are needed. The no-reference/blind image
quality assessment (NR IQA) refers to quality assessment of images by an algorithm where
only the distorted image is accessible and no information about the reference image is avail-
able. In this article, several FR IQA and NR IQA tools are used to evaluate the performance of
diferent upsampling methods. The quality metrics are introduced in the following section:
2.1 FR IQA Tools
2.1.1 Peak signal-to-noise ratio
PSNR is one of the most prevalent tools for image quality evaluation defined by the following
equation: