Professional Documents
Culture Documents
Wang 2014
Wang 2014
Abstract—Image quality evaluation is very important. In appli- image and the original image [1]–[7]. The widely used quality
cations involving signal transmission, the Reduced- or No-Refer- metrics in this category are the PSNR, wPSNR, Watson JND,
ence quality metrics are generally more practical than the Full- SSIM, etc. The Full-Reference quality metrics provide more
Reference metrics. In this study, we propose a quality estimation
method based on a novel semi-fragile and adaptive watermarking
accurate quality evaluation results comparing to the Reduced
scheme. The proposed scheme uses the embedded watermark to es- or No-Reference quality metrics. However, the Full-Reference
timate the degradation of cover image under different distortions. quality metrics become less practical when the original image is
The watermarking process is implemented in DWT domain of the not available. The Reduced-Reference quality metrics evaluate
cover image. The correlated DWT coefficients across the DWT sub- the quality of a distorted image using partial information of the
bands are categorized into Set Partitioning in Hierarchical Trees original image. In literature, such partial information can be
(SPIHT). Those SPHIT trees are further decomposed into a set of
bitplanes. The watermark is embedded into the selected bitplanes some features extracted from the original image [8]–[10]. The
of the selected DWT coefficients of the selected tree without causing Reduced-Reference quality metrics do not require the presence
significant fidelity loss to the cover image. The accuracy of the of the original image for quality evaluation. However, the
quality estimation is made to approach that of Full-Reference met- partial information of the original image need to be transmitted
rics by referring to an "Ideal Mapping Curve" computed a priori. to the receiver side either through an ancillary channel or by
The experimental results show that the proposed scheme can esti-
embedding into the transmitted image. The sacrifice of band-
mate image quality in terms of PSNR, wPSNR, JND and SSIM with
high accuracy under JPEG compression, JPEG2000 compression, width for transmitting the additional information needs to be
Gaussian low-pass filtering and Gaussian noise distortion. The re- considered. The No-Reference quality metrics estimate image
sults also show that the proposed scheme has good computational quality without accessing the original image [11]–[16]. In
efficiency for practical applications. practical applications, different quality metrics can measure the
Index Terms—Bitplane decomposition, complexity analysis, image degradation from different angle. For example, PSNR
DWT based watermark embedding, HVS masking, SPIHT tree measures image quality mathematically in terms of MSE; JND
structure, watermarking based image quality estimation. and SSIM intend to measure image quality with more emphasis
on the perceptual experience. Multiple metrics are often used
in practical applications to give a full coverage of the image
I. INTRODUCTION quality evaluation. In terms of applicability in signal transmis-
sion, Reduced- or No-Reference metrics are more preferable
1520-9210 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
312 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 16, NO. 2, FEBRUARY 2014
it is hard to examine which kind of image features are suit- across the subbands in the wavelet decomposed image. The
able for providing accurate quality estimation. The image-fea- DWT and SPIHT together provide a good summarization of
ture-independent schemes simplify the situation [19]–[23]. In local region characteristics of an image which is important for
these schemes, the watermark is independent of the image fea- adaptive watermark embedding. In this proposed scheme, all the
tures and needs to be known at the receiver side. The water- correlated DWT coefficients across the subbands are grouped
mark degradation is evaluated using the distorted watermark together using the SPIHT tree structure. The DWT decomposed
and the original watermark. One challenging task to develop the image is further decomposed into a set of bitplane images. In this
image-feature-independent watermarking based quality metric case, each DWT coefficient is decomposed into a sequence of
is to enable the embedded watermark to accurately reflect the binary bits. The binary watermark bits are embedded into the se-
quality changes of cover image under distortions. This requires lected bitplanes of the selected DWT coefficients of the selected
the watermark being adaptively embedded in cover images ac- trees. The HVS masking is used to guide the bitplane selection.
cording to the characteristics of cover images, so that the em- As found in the experiments, the higher frequency DWT sub-
bedded watermark degrades in a similar way as the cover images bands and less significant bitplanes are more sensitive to dis-
under distortions. This is a critical part of the whole scheme and tortions, and vice versa [19]. Therefore, the robustness of the
directly affects the accuracy of quality estimation. watermark is controlled by two factors:
In [19], an image-feature-independent watermarking based (a) The percentages of the watermark bits embedded into the
quality estimation scheme was proposed which attempts to three DWT levels, respectively, and,
achieve the quality estimation accuracy of the Full-Reference (b) The selection of bitplanes for watermark embedding.
objective metrics. In the scheme, the watermark degradation Thus, for different selected trees, the watermark embedding
is measured using the True Detection Rates (TDR). Then the strengths are different.
image quality is estimated by mapping the calculated TDR to The proposed scheme is tested in terms of PSNR, wPSNR,
a quality value using an empirical mapping function which Watson JND and SSIM, and under JPEG compression,
is experimentally generated and is used as a priori at the JPEG2000 compression, Gaussian low-pass filtering and
receiver side. An iterative process is used to find the optimal Gaussian noise addition. The results show the effectiveness of
watermark embedding strength by experimentally testing the the proposed scheme.
image degradation characteristics so that the quality estimation The rest of the paper is organized as follows. Section II,
error can be minimized. This iterative process provides high Section III and Section IV present the proposed watermark
accuracy to the quality estimation. However, it also introduces embedding scheme in detail. Section V describes the watermark
relatively high computational complexity which makes it less extraction and quality evaluation scheme. Section VI details
suitable for certain applications. Meanwhile, in the scheme, the generation of the “Ideal Mapping Curve”. Section VII
the human perception characteristics have not been taken into shows the experimental results and accuracy evaluations of
consideration during the watermark embedding process and the proposed scheme. Section VIII lists the advantages of the
the quality of the watermarked image is about 40 dB in PSNR proposed scheme. Section IX concludes the paper and discusses
on average. Thus, the goal of our research in this paper is future work.
to keep the accuracy of quality estimation achieved in [19]
while improving the computational efficiency and reducing the II. THE PROPOSED WATERMARK EMBEDDING SCHEME
image quality degradation caused by the watermark embedding
The proposed watermark embedding scheme is shown in
process. Here, the term “accuracy” evaluates the correlation of
Fig. 1, which includes the watermark pre-processing, the image
the estimated quality and the quality calculated using the ex-
isting objective Full-Reference quality metrics, such as PSNR. pre-analysis, and the watermark embedding. The watermark
The closer the estimated quality to the calculated quality, the embedding process consists of the following three steps:
more accurate the quality estimation, and vice versa. (i) Apply 3-level DWT to the original image to obtain the
In this paper, we propose a new approach that well meets DWT decomposed image. The 3-level DWT decomposed
our research goal addressed above. In the proposed scheme, the subbands are denoted as shown in Fig. 2(a). These denota-
adaptive watermark embedding strength is estimated by ana- tions for the 10 DWT decomposed subbands will be used
lyzing the quality degradation characteristics of the cover image throughout the paper.
and no iterative adjustment loops are used, which significantly (ii) Embed the watermark with adaptive embedding strength
improves the computational efficiency and theoretically makes using the tree structure based watermark embedder which
the proposed scheme applicable to real-time video quality es- will be presented in Section III in detail. The output of the
timation. Moreover, the strategies including the HVS masking watermark embedder is the watermarked DWT image.
are used to guide the watermark embedding process. With the (iii) Apply 3-level inverse DWT to the watermarked DWT
proposed scheme, the quality of the watermarked images refer- image to obtain the watermarked image.
ring to the original images is about 48 dB in PSNR on average, The watermark pre-processing procedure is shown in the
which is an 8-dB improvement over the scheme in [19]. grey-dashed block with number 1 on its upper right corner. In
The proposed scheme is based on adaptive watermarking and this procedure, a two dimensional original watermark is orga-
tree structure in the DWT domain. Recently, the Set Partitioning nized column by column into a one dimensional sequence. We
in Hierarchical Trees (SPIHT) has become one of the most pop- denote the length of the original watermark sequence as len. To
ular image and video coding method because of its efficiency increase the probability of the correct watermark bit extraction
which is accomplished by exploiting the inherent similarities at the receiver side, every bit in the original watermark sequence
WANG et al.: ADAPTIVE WATERMARKING AND TREE STRUCTURE BASED IMAGE QUALITY ESTIMATION 313
(b) With the group indices, the watermark bits are assigned to where
the images using (5).
if
o.w.
if
if
(5) if
for HL blocks
for HH blocks
for LH blocks
and
(b) Background luminance: Intensity variations are less vis-
ible over the brighter and darker areas. The luminance
when
masking is denoted as .
when (6)
when
Fig. 6. Illustration of the generated HVS masks. (a) The original image Bar-
bara. (b) The nine HVS masks generated for the nine DWT detail blocks.
(13)
Fig. 11. Examples of the generation of “ideal mapping curves” under Gaussian
low-pass filtering.
TABLE I
RANGE OF DISTORTION STRENGTHS AND CORRESPONDING
WATERMARKED IMAGE QUALITIES IN PSNR
the quality of the distorted images is estimated in terms of D. The Empirical Grouping Thresholds
PSNR, wPSNR, Watson JND [29] and SSIM [2], respectively. For each set of experiments, the grouping thresholds used in
Therefore, totally 16 sets of experiments are conducted. In (4) are empirically set by testing the first 50 images in the image
each set of the experiments, 100 images are tested. To assess library using the following steps:
the performance of the proposed scheme, the quality of the (a) Based on the research done in [19], empirically set up the
distorted image comparing to the original image is calculated watermark bit assignment, , as listed in (5).
as well. With both the estimated quality and calculated quality, (b) With each , test the quality estimation error using the
the quality estimation accuracy is evaluated using MAE (Mean 50 images.
Absolute Error), Pearson correlation coefficient, and RMSE (c) For each test image, find the minimum quality estimation
(Root Mean Squared Error) respectively. In the experiments, error and the corresponding optimal .
the image quality degradation caused by both the watermark (d) Calculate the content complexity values for the 50 test
embedding and the distortion is considered in the performance images using the local intensity difference threshold 0.17,
evaluation. where the maximum intensity value for the images is 1.
(e) Referring to the optimal assigned to the 50 test im-
C. Choosing the Range for the Distortion Strength ages, categorize the calculated content complexity values
The watermarked images are distorted using a set of dis- into 6 groups.
tortion strengths varying in a selected range, , (f) Set up the empirical thresholds by examining the catego-
where is the distortion strength; and respectively are rized content complexity values.
the lowest and the highest distortion strength used in the ex- For different set of experiments, the empirical grouping
periments. We choose the lowest distortion strength available thresholds may be different. The thresholds used in the experi-
in Matlab as . For example, under JPEG compression, the ments are listed in Table II. As found in the experiments, under
quality factor controls the distortion strength and . Gaussian noise distortion with a fixed noise standard deviation,
is so set that the average quality of the watermarked images dis- the quality of different distorted images varies slightly no
torted using , , is in the range of [24.50, 25.50] dB matter how complex the images are. Thus, under Gaussian
in PSNR. , where the noise distortion, we use the grouping thresholds the same as
is the watermarked image and is the original image. On those used under JPEG compression for convenience. These
the other hand, when the quality of the distorted image is below thresholds are stored at the receiver side once they are tested.
30 dB in PSNR, the image quality is poor. The lower the image
quality, the more meaningless to focus on the quality estimation E. Experimental Results Under Different Distortions
accuracy. Thus, we choose another high boundary for : For the four types of distortions, the distortion strengths used
if in PSNR. The 150 images in in the experiments are listed in the following:
the image library are used to choose , and . The range • JPEG: ;
of the distortion strengths for the four types of distortions used • JPEG2000:
in the experiments are listed in the following table: .
WANG et al.: ADAPTIVE WATERMARKING AND TREE STRUCTURE BASED IMAGE QUALITY ESTIMATION 321
Fig. 13. The generated “ideal mapping curves” in terms of different full-reference quality metrics under different distortions.
TABLE II
THE EMPIRICAL THRESHOLDS USED FOR THE 16 SETS OF EXPERIMENTS , the JND value increases rapidly with the decreasing TDR
value. When TDR decreases from 0.4 to 0.31, the JND value in-
creases from 8 to 40. Thus, we need to use a TDR value varying
in a 0.09 value range to estimate image quality varying in a 32
JND value range. In this case, a small variation of the TDR value
may result in a relatively significant quality estimation error. So,
• Low-pass filtering: we can predict that this rapid increase of JND will greatly reduce
. the accuracy of the quality estimation when . For
• Gaussian noise distortion: additional observations, the quality estimation accuracy when
. the calculated quality is smaller than 8 JND is evaluated as well.
For the 16 sets of experiments, the 16 generated “Ideal Similar analysis can be made to Fig. 13(g), (k), and (o).
Mapping Curves” are shown in Fig. 13(a)–(p). For example, Fig. 13(d), (h), (l) and (p) are the “Ideal Mapping Curves”
Fig. 13(a), (e), (i) and (m) show the “Ideal Mapping Curves” used to estimate image quality in terms of SSIM. In Fig. 13(d),
for the quality estimation in terms of PSNR. With increasing the left half of the curve has a very big slope. When the TDR
distortion strength, the calculated PSNR value and the calcu- decreases from 0.42 to 0.31, the calculated SSIM changes from
lated TDR value decrease monotonically. 0.9871 to 0.7. The SSIM values drop rapidly corresponding a
Fig. 13(c), (g), (k) and (o) are the “Ideal Mapping Curves” small change of TDR value. It is straightforward that a small
used to estimate image quality in terms of JND. With increasing variation of the TDR value may cause relatively bigger estima-
distortion strength, the calculated TDR decreases and the cal- tion error and the divergence of the quality points referring to
culated JND value increases. When TDR equals to 1, the JND the solid matching line may be magnified.
is 0. In Fig. 13(c), is a slope change point and The 16 sets of experimental results of image quality estima-
JND, where is the quality. When tion tested with distortion strengths are shown in
322 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 16, NO. 2, FEBRUARY 2014
Fig. 14. The experimental results tested with distortions in terms of different Full-Reference quality metrics under different distortions.
Fig. 14(a)–(p). In the figures, the x-axis is the estimated quality. scattering. The poorer the image quality, the more meaning-
The y-axis is the correspondingly calculated quality. The black less to focus on the quality estimation accuracy. The accuracy
points are the quality points achieved under different distortion of the image quality estimation evaluated using MAE, Pearson
strengths by testing the 100 images. The solid line is the quality correlation coefficient and RMSE is summarized in
matching line with a slope of 1, which indicates that the esti- Table III, Table IV and Table V, respectively. In the tables,
mated quality equals to the calculated quality. The discrepancy and respectively indicate the range of the distortion strengths
from the solid line to the quality points indicate the accuracy of and , which are described in Section VII-C and
the quality estimation. The more convergent the quality points represents stronger distortion strength than . LPF stands
to the solid line, the more accurate the quality estimation and for low passing filtering, and Gaussian stands for Gaussian noise
vice versa. If all the quality points fall on the solid line, the es- addition.
timated quality equals to the calculated quality and the highest From Table III through Table V, we can see that:
quality estimation accuracy is achieved. For each sub-figure in (a) The proposed scheme has good accuracy to estimate
Fig. 14, the corresponding quality estimation accuracy in MAE image quality in terms of PSNR and wPSNR. Under
is listed. Recall that in PSNR. When the the four mentioned distortions, with both the distortion
quality of a distorted image is lower than 30 dB in PSNR, the strengths and , the calculated for
image quality becomes poor and the quality points become more the quality estimation in PSNR are respectively lower
WANG et al.: ADAPTIVE WATERMARKING AND TREE STRUCTURE BASED IMAGE QUALITY ESTIMATION 323
TABLE IV
SUMMARY OF THE ACCURACY IN PEARSON CORRELATION In the proposed scheme, four sets of parameters and thresh-
olds have been introduced: the grouping thresholds for calcu-
lating in (4), the watermark bit assignment parameters
in (5), the parameters used to calculate the HVS mask
in (12), and the mapping parameters from HVS masks to
bitplane indices in (14). The first set of parameters in (4) is
image independent and distortion dependent. The rest three sets
of empirical parameters are image and distortion independent,
TABLE V which means that they are fixed for all images and all types of
SUMMARY OF THE ACCURACY IN RMSE
distortions. All these parameters and thresholds work together
to balance the robustness of algorithm and the quality degrada-
tion caused by the watermark embedding.
quality metrics, such as PSNR, wPSNR, JND and SSIM. Thus, [15] H. Wu and M. Yuen, “A generalized block edge impairment metric for
at the receiver side of a communication system, without the orig- video coding,” Signal Process. Lett., vol. 4, no. 11, pp. 317–320, 1997.
[16] F. Battisti, M. Carli, and A. Neri, “No-reference quality metric for color
inal image, the quality of a distorted image can still be assessed. video communication,” in Proc. Int. Workshop Video Processing and
Based on the tree structure, the binary watermark is em- Quality Metrics, 2012, pp. 1–6.
bedded into the selected bitplanes of the selected DWT [17] A. Bhattacharya, S. Palit, N. Chatterjee, and G. Roy, “Blind assessment
of image quality employing fragile watermarking,” in Proc. IEEE 7th
coefficients with adaptive watermark embedding strength. Int. Symp. ISPA, 2011, pp. 431–436.
The watermark embedding strength is assigned to an image [18] S. Altous, M. K. Samee, and J. Gotze, “Reduced reference image
by pre-analyzing its content complexity in the spatial domain quality assessment for jpeg distortion,” in Proc. IEEE 7th Int. Symp.
ELMAR, 2011, pp. 97–100.
and the perceptual masking effect of the DWT decomposed [19] S. Wang, D. Zheng, J. Zhao, W. J. Tam, and F. Speranza, “An image
image in the DWT domain. Meanwhile, the watermark is not quality evaluation method based on digital watermarking,” IEEE
embedded in the approximation subband, which reduces loss Trans. Circuits Syst. Video Technol., vol. 17, no. 1, pp. 98–105, 2007.
[20] D. Zheng, J. Zhao, W. Tam, and F. Speranza, “Image quality mea-
in image quality caused by embedding the watermark. The surement by using digital watermarking,” in Proc. IEEE Int. Workshop
experimental results show that the proposed scheme works Haptic, Audio and Visual Environments and Their Applications, 2003,
effectively. pp. 65–70.
[21] M. Farias, M. Carli, and S. Mitra, “Objective video quality metric based
In future work, the proposed scheme will be further devel- on data hiding,” IEEE Trans. Consum. Electron., vol. 51, no. 3, pp.
oped to estimate the quality of an image distorted by multiple 983–992, 2005.
distortions. Meanwhile, experiments about image quality esti- [22] E. Nezhadarya, Z. Wang, and R. Ward, “Image quality monitoring
using spread spectrum watermarking,” in Proc. IEEE Int. Conf. Image
mation in terms of subjective quality scores will be conducted. Processing, 2009, pp. 2233–2236.
Since the proposed scheme has good computational efficiency, [23] A. Mishra, A. Jain, M. Narwaria, and C. Agarwal, “An experimental
it is feasible to further develop the proposed scheme for video study into objective quality assessment of watermarked images,” Int.
J. Image Process., vol. 5, no. 2, pp. 199–219, 2011.
quality evaluation. [24] D. Taubman and M. W. Marcellin, JPEG2000: Image Compres-
sion Fundamentals, Standards and Practice. Norwell, MA, USA:
Kluwer, 2002.
[25] A. Said and W. Pearlman, “A new, fast and efficient image codec based
on set partitioning in hierarchical trees,” IEEE Trans. Circuits Syst.
REFERENCES Video Technol., vol. 6, no. 3, pp. 243–250, 1996.
[26] M. Jamzad and F. Yaghmaee, “Achieving higher stability in water-
[1] I. J. Cox, M. L. Miller, J. A. Bloom, J. Fridrich, and T. Kalker, Digital marking according to image complexity,” Scientia Iranica J., vol. 13,
Watermarking and Steganography. San Mateo, CA, USA: Morgan no. 4, pp. 404–412, 2006.
Kaufmann, 2008. [27] M. Barni, “Improved wavelet-based watermarking through pixel-wise
[2] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image masking,” IEEE Trans. Image Process., vol. 10, no. 5, pp. 783–791,
quality assessment: From error visibility to structural similarity,” IEEE 2001.
Trans. Image Process., vol. 13, no. 4, pp. 600–612, 2004. [28] A. Lewis and G. Knowles, “Image compression using the 2-D wavelet
[3] S. Wolf, M. Pinson, S. Voran, and A. Webster, “Objective quality as- transform,” IEEE Trans. Image Process., vol. 1, no. 52, pp. 244–250,
sessment of digitally transmitted video,” in Proc. IEEE Pacific Rim 1992.
Conf. Communications, Computers and Signal Processing, 1991, vol. [29] [Online]. Available: http://vision.arc.nasa.gov/dctune/(last visited in
2, pp. 477–482. Sep. 2011).
[4] A. Watson, Q. Hu, and J. McGowan, “Digital video quality metric [30] D. Sheskin, Handbook of Parametric and Nonparametric Statistical
based on human vision,” J. Electron. Imag., vol. 10, pp. 20–29, 2001. Procedures, 2nd ed. London, U.K.: Chapman & Hall CRC, 2000.
[5] A. Shnayderman, A. Gusev, and A. Eskicioglu, “An SVD-based
grayscale image quality measure for local and global assessment,”
IEEE Trans. Image Process., vol. 15, no. 2, pp. 422–429, 2006.
[6] G. Zhai, W. Zhang, X. Yang, S. Yao, and Y. Xu, “GES: A new image
quality assessment metric based on energy features in Gabor transform
domain,” in Proc. IEEE Int. Symp. Circuits and Systems, 2006, pp. Sha Wang received her B.Eng in thermal power
1715–1718. engineering from North China Electric Power Uni-
[7] M. Carnec, P. Callet, and D. Barba, “Full reference and reduced ref- versity, China in 2002, and received her Master’s
erence metrics for image quality assessment,” in Proc. 7th Int. Symp. and Ph.D. degree in electrical engineering from
Signal Processing and Its Applications, 2003, vol. 1, pp. 477–480. University of Ottawa, Canada, respectively in 2005
[8] Q. Li and Z. Wang, “Reduced-reference image quality assessment and 2013. Her research interests are on image and
using divisive normalization-based image representation,” IEEE J. video processing.
Select. Topics Signal Process., vol. 3, no. 2, pp. 201–211, 2009.
[9] Z. Wang, G. Wu, H. R. Sheikh, E. P. Simoncelli, E. H. Yang, and A. C.
Bovik, “Quality-aware images,” IEEE Trans. Image Process., vol. 15,
no. 6, pp. 1680–1689, 2006.
[10] T. Kusuma, H. Zepernick, and M. Caldera, “On the development of a
reduced-reference perceptual image quality metric,” in Proc. Systems
Communications, 2005, pp. 178–184.
[11] P. Lecallet, C. Viard-Gaudin, S. Pechard, and E. Caillault, “No refer- Dong Zheng received his B.E. degree in electrical
ence and reduced reference video quality metrics for end to end QoS engineering from Wuhan University, China in 1996
monitoring,” IEICE Trans. Commun., vol. E85-B, no. 2, pp. 289–296, and received his M.A.Sc and Ph.D. in electrical engi-
2006. neering from University of Ottawa, Canada respec-
[12] D. Turaga, Y. Chen, and J. Caviedes, “No reference PSNR estimation tively in 2003 and 2008. He worked as a senior re-
for compressed pictures,” Signal Process.: Image Commun., vol. 19, search engineer in Dilithiumnetwork Inc., California
no. 2, pp. 173–184, 2004. from 2007 to 2008. He worked as a research scientist
[13] P. Marziliano, F. Dufaux, S. Winkler, and T. Ebrahimi, “A no-reference at Communication Research Centre of Canada from
perceptual blur metric,” in Proc. IEEE Int. Conf. Image Processing, 2008 to 2012. Since 2012, he has been working as
2002, vol. 3, pp. 57–60. a staff software engineer of Magnum Semiconductor
[14] W. Ci, H. Dong, Z. Wu, and Y. Tan, “Example-based objective quality on video processing. His current research interests in-
estimation for compressed images,” IEEE Multimedia, to be published. clude digital video and image coding and quality assessment.
WANG et al.: ADAPTIVE WATERMARKING AND TREE STRUCTURE BASED IMAGE QUALITY ESTIMATION 325
Jiying Zhao (M’00) received the Ph.D. degree Filippo Speranza received his Ph.D. in Experi-
in electrical engineering from North China Elec- mental Psychology from the University of Toronto
tric Power University, and the Ph.D. degree (Canada) in 1999. From 2001 to 2013, he worked
in computer engineering from Keio University. as Research Scientist with Industry Canada—Com-
He is a professor with the School of Electrical munications Research Centre (CRC), where he
Engineering and Computer Science, University of investigated stereoscopic imaging and subjective
Ottawa, Ottawa, Canada. His research interests picture quality assessment techniques and methods.
include image and video processing and multi- He is an adjunct professor at the University of
media communications. Dr. Zhao is a member Ottawa.
of the Institute of Electronics, Information and
Communication Engineers (IEICE) and is also a
member of the Professional Engineers Ontario (PEO).