Image Processing Reference
In-Depth Information
Fig. 9.1 The
Orbi
sequence (a) depth map, (b) extracted binary edge mask of the depth map using
Sobel
filtering, (c) corresponding color image, and (d) extracted binary edge mask of the color
image using
Sobel
filtering
9.2 Proposed Real-Time Quality Evaluation Strategy
The color image and the corresponding depth map share common object boundaries
even though individual pixel values are different (see Fig.
9.1
). The gradient/edge
information of these images can be used to detect object boundaries/contours.
The edge information of the depth map represents depth map object boundaries
and different depth levels (see Fig.
9.1b
). Depth discontinuities occur around these
edges; hence, their preservation is vital in 3D video rendering based on the DIBR
method [
29
], since any imperfections around these edges will severely affect the
quality of the rendered left and right views. The effectiveness of using edge
information for measuring the depth map quality is shown in [
27
,
28
]. Similarly
to the depth edge characteristics, gradient information of the color image sequence
also identifies object contours and boundaries (see Fig.
9.1d
) and is also used in
color image quality evaluation [
25
,
26
]. Figures
9.2
and
9.3
demonstrate the
effectiveness of using gradient information to represent artifacts of the images
due to compression and packet losses occurring during transmission. In these
figures, both 3D sequences are encoded using the Quantization Parameter (
QP
)
ΒΌ
30 and corrupted with 20 %
PLR
[
30
]. This shows that edge information can be
