Characteristic measurements for the qualification of reflection scanners in the evaluation of image quality attributes
Author(s):
Eric K. Zeise
Show Abstract
The claimed specifications of reflection scanners make their utilization for analytic measurement very tempting. This
paper summarizes an effort by Working Group 4 of ISO/IEC JTC-1 SC28 to develop evaluation methods that can be
used to characterize the performance of reflection scanners with the goal of developing a sufficient characterization set
to serve as evaluation methods in conformance testing for future image quality standards. Promising evaluation methods
for tone-scale, spatial and temporal uniformity, spatial distortion, SNR, dynamic range and flare characteristics will be
described.
W1.1 macro uniformity
Author(s):
D. René Rasmussen;
Frans Gaykema;
Yee S. Ng;
Kevin D. Donohue;
William C. Kress;
Susan Zoltner
Show Abstract
The INCITS W1.1 macro-uniformity team works towards the development of a standard for evaluation of perceptual
image quality of color printers. The team specifically addresses the types of defects that fall in the category of macrouniformity,
such as streaks, bands and mottle. This paper provides a brief summary of the status of this work, and
describes recent results regarding the precision of the macro-uniformity quality ruler for assessment of typical printer
defects.
Measurement of contributing attributes of perceived printer resolution.
Author(s):
Eric K. Zeise;
Sang Ho Kim;
Brian E. Cooper;
Franz Sigg
Show Abstract
Several measurable image quality attributes contribute to the perceived resolution of a printing system. These
contributing attributes include addressability, sharpness, raggedness, spot size, and detail rendition capability. This
paper summarizes the development of evaluation methods that will become the basis of ISO 29112, a standard for the
objective measurement of monochrome printer resolution.
Softcopy quality ruler method: implementation and validation
Author(s):
Elaine W Jin;
Brian W. Keelan;
Junqing Chen;
Jonathan B. Phillips;
Ying Chen
Show Abstract
A softcopy quality ruler method was implemented for the International Imaging Industry Association (I3A) Camera
Phone Image Quality (CPIQ) Initiative. This work extends ISO 20462 Part 3 by virtue of creating reference digital
images of known subjective image quality, complimenting the hardcopy Standard Reference Stimuli (SRS). The
softcopy ruler method was developed using images from a Canon EOS 1Ds Mark II D-SLR digital still camera (DSC)
and a Kodak P880 point-and-shoot DSC. Images were viewed on an Apple 30in Cinema Display at a viewing distance of
34 inches. Ruler images were made for 16 scenes. Thirty ruler images were generated for each scene, representing ISO
20462 Standard Quality Scale (SQS) values of approximately 2 to 31 at an increment of one just noticeable difference
(JND) by adjusting the system modulation transfer function (MTF). A Matlab GUI was developed to display the ruler
and test images side-by-side with a user-adjustable ruler level controlled by a slider. A validation study was performed at
Kodak, Vista Point Technology, and Aptina Imaging in which all three companies set up a similar viewing lab to run the
softcopy ruler method. The results show that the three sets of data are in reasonable agreement with each other, with the
differences within the range expected from observer variability. Compared to previous implementations of the quality
ruler, the slider-based user interface allows approximately 2x faster assessments with 21.6% better precision.
Correlating objective and subjective evaluation of texture appearance with applications to camera phone imaging
Author(s):
Jonathan B. Phillips;
Stephen M. Coppola;
Elaine W. Jin;
Ying Chen;
James H. Clark;
Timothy A. Mauer
Show Abstract
Texture appearance is an important component of photographic image quality as well as object recognition. Noise
cleaning algorithms are used to decrease sensor noise of digital images, but can hinder texture elements in the process.
The Camera Phone Image Quality (CPIQ) initiative of the International Imaging Industry Association (I3A) is
developing metrics to quantify texture appearance. Objective and subjective experimental results of the texture metric
development are presented in this paper. Eight levels of noise cleaning were applied to ten photographic scenes that
included texture elements such as faces, landscapes, architecture, and foliage. Four companies (Aptina Imaging, LLC,
Hewlett-Packard, Eastman Kodak Company, and Vista Point Technologies) have performed psychophysical evaluations
of overall image quality using one of two methods of evaluation. Both methods presented paired comparisons of images
on thin film transistor liquid crystal displays (TFT-LCD), but the display pixel pitch and viewing distance differed. CPIQ
has also been developing objective texture metrics and targets that were used to analyze the same eight levels of noise
cleaning. The correlation of the subjective and objective test results indicates that texture perception can be modeled
with an objective metric. The two methods of psychophysical evaluation exhibited high correlation despite the
differences in methodology.
Imaging performance taxonomy
Author(s):
Don Williams;
Peter D. Burns;
Larry Scarff
Show Abstract
A significant challenge in the adoption of today's digital imaging standards is a clear connection to how they relate to
today's vernacular digital imaging vocabulary. Commonly used terms like resolution, dynamic range, delta E, white
balance, exposure, or depth of focus are mistakenly considered measurements in their own right and are frequently
depicted as a disconnected shopping list of individual metrics with little common foundation. In fact many of these are
simple summary measures derived from more fundamental imaging science/engineering metrics, adopted in existing
standard protocols.
Four important underlying imaging performance metrics are; Spatial Frequency Response (SFR), Opto-Electronic
Conversion Function (OECF), Noise Power Spectrum (NPS), and Spatial Distortion. We propose an imaging
performance taxonomy. With a primary focus on image capture performance, our objective is to indicate connections
between related imaging characteristics, and provides context for the array of commonly used terms. Starting with the
concepts of Signal and Noise, the above imaging performance metrics are related to several simple measures that are
compatible with testing for design verification, manufacturing quality assurance, and technology selection evaluation.
Extended use of ISO 15739 incremental signal-to-noise ratio as reliability criterion for multiple-slope wide dynamic range image capture
Author(s):
Dirk Hertel
Show Abstract
In the emerging field of automotive vision, video capture is the critical front-end of driver assistance and active safety
systems. Previous photospace measurements have shown that light levels in natural traffic scenes may contain an
extremely wide intra-scene intensity range. This requires the camera to have a wide dynamic range (WDR) for it to adapt
quickly to changing lighting conditions and to reliably capture all scene detail.
Multiple-slope CMOS technology offers a cost-effective way of adaptively extending dynamic range by partially
resetting (recharging) the CMOS pixel once or more often within each frame time. This avoids saturation and leads to a
response curve with piecewise linear slopes of progressively increasing compression.
It was observed that the image quality from multiple-slope image capture is strongly dependent on the control (height
and time) of each reset barrier. As compression and thus dynamic range increase there is a trade-off against contrast and
detail loss.
Incremental signal-to-noise ratio (iSNR) is proposed in ISO 15739 for determining dynamic range. Measurements and
computer simulations revealed that the observed trade-off between WDR extension and the loss of local detail could be
explained by a drop in iSNR at each reset point. If a reset barrier is not optimally placed then iSNR may drop below the
detection limit so that an 'iSNR hole' appears in the dynamic range. Thus ISO 15739 iSNR has gained extended utility:
it not only measures the dynamic range limits but also defines dynamic range as the intensity range where detail
detection is reliable. It has become a critical criterion when designing adaptive barrier control algorithms that maximize
dynamic range while maintaining the minimum necessary level of detection reliability.
Web-based psychometric evaluation of image quality
Author(s):
Iris Sprow;
Zofia Baranczuk;
Tobias Stamm;
Peter Zolliker
Show Abstract
The measurement of image quality requires the judgement by the human visual system. This paper describes
a psycho-visual test technique that uses the internet as a test platform to identify image quality in a more
time-effective manner, comparing the visual response data with the results from the same test in a lab-based
environment and estimate the usefulness of the internet as a platform for scaling studies.
Development of a balanced test image for visual print quality evaluation
Author(s):
Hanne Salmi;
Raisa Halonen;
Tuomas Leisti;
Pirkko Oittinen;
Hannu Saarelma
Show Abstract
A test image for color still image processes was developed. The image is based on general requirements on the content
and specific requirements arising from the quality attributes of interest. The quality attributes addressed in the study
include sharpness, noise, contrast, colorfulness and gloss. These were chosen based on visual relevance in studies of the
influence of paper in digital printing. Further requirements such as arising from the use cases of the image are discussed
based on eye tracking data and self-report of the usefulness of different objects for quality evaluation. From the
standpoint of being sufficiently sensitive to quality variations of the imaging systems to be measured the reference test
image needs to represent quality maxima in terms of the relevant quality parameters. As for different viewing times, no
object should be exceedingly salient. The paper presents the procedure of developing the test image and discusses its
merits and shortcomings from the standpoint of future development.
Perceptual image attribute scales derived from overall image quality assessments
Author(s):
Kyung Hoon Oh;
Sophie Triantaphillidou;
Ralph E. Jacobson
Show Abstract
Psychophysical scaling is commonly based on the assumption that the overall quality of images is based on the
assessment of individual attributes which the observer is able to recognise and separate, i.e. sharpness, contrast, etc.
However, the assessment of individual attributes is a subject of debate, since they are unlikely to be independent from
each other.
This paper presents an experiment that was carried to derive individual perceptual attribute interval scales from overall
image quality assessments, therefore examine the weight of each individual attribute to the overall perceived quality. A
psychophysical experiment was taken by fourteen observers. Thirty two original images were manipulated by adjusting
three physical parameters that altered image blur, noise and contrast. The data were then arranged by permutation, where
ratings for each individual attribute were averaged to examine the variation of ratings in other attributes.
The results confirmed that one JND of added noise and one JND of added blurring reduced image quality more than did
one JND in contrast change. Furthermore, they indicated that the range of distortion that was introduced by blurring
covered the entire image quality scale but the ranges of added noise and contrast adjustments were too small for
investigating the consequences in the full range of image quality. There were several interesting tradeoffs between noise,
blur and changes in contrast. Further work on the effect of (test) scene content was carried out to objectively reveal
which types of scenes were significantly affected by changes in each attribute.
Subjective experience of image quality: attributes, definitions, and decision making of subjective image quality
Author(s):
Tuomas Leisti;
Jenni Radun;
Toni Virtanen;
Raisa Halonen;
Göte Nyman
Show Abstract
Subjective quality rating does not reflect the properties of the image directly, but it is the outcome of a quality decision
making process, which includes quantification of subjective quality experience. Such a rich subjective content is often
ignored. We conducted two experiments (with 28 and 20 observers), in order to study the effect of paper grade on image
quality experience of the ink-jet prints. Image quality experience was studied using a grouping task and a quality rating
task. Both tasks included an interview, but in the latter task we examined the relations of different subjective attributes in
this experience. We found out that the observers use an attribute hierarchy, where the high-level attributes are more
experiential, general and abstract, while low-level attributes are more detailed and concrete. This may reflect the
hierarchy of the human visual system. We also noticed that while the observers show variable subjective criteria for IQ,
the reliability of average subjective estimates is high: when two different observer groups estimated the same images in
the two experiments, correlations between the mean ratings were between .986 and .994, depending on the image
content.
Toward an automatic subjective image quality assessment system
Author(s):
M. Chambah;
S. Ouni;
M. Herbin;
E. Zagrouba
Show Abstract
Usually in the field of image quality assessment the terms "automatic" and "subjective" are often incompatible. In fact,
when it comes to image quality assessment, we have mostly two kinds of evaluation techniques: subjective evaluation
and objective evaluation. Only objective evaluation techniques being automatizable, while subjective evaluation
techniques are performed by a series of visual assessment done by expert or non-expert observers.
In this paper, we will present a first attempt to an automatic subjective quality assessment system. The system computes
some perception correlated color metrics from a learning set of images. During the learning stage a subjective assessment
by users is required so that the system matches the subjective opinions with computed metrics on a variety of images.
Once the learning process is over, the system operates in an automatic mode using only the learned knowledge and the
reference free computed metrics from the images to assess. Results and also future prospects of this work are presented.
Methods for measuring display defects as correlated to human perception
Author(s):
H. Kostal;
G. Pedeville;
R. Rykowski
Show Abstract
Human vision and perception are the ultimate determinants of display quality, however human judgment is variable,
making it difficult to define and apply quantitatively in research or production environments. However, traditional
methods for automated defect detection do not relate directly to human perception - which is especially an issue in
identifying just noticeable differences. Accurately correlating human perceptions of defects with the information that can
be gathered using imaging colorimeters offers an opportunity for objective and repeatable detection and quantification of
such defects. By applying algorithms for just noticeable differences (JND) image analysis, a means of automated,
repeatable, display analysis directly correlated with human perception can be realized. The implementation of this
technique and typical results are presented. Initial application of the JND analysis provides quantitative information that
allows a quantitative grading of display image quality for FPDs and projection displays, supplementing other defect
detection techniques.
A strobe-based inspection system for drops-in-flight
Author(s):
Yair Kipman;
Prashant Mehta;
Kate Johnson
Show Abstract
Imaging and measurement of drops-in-flight often relies on the measurement system's ability to drive the print head
directly in order to synchronize the strobe for repeatable image capture. In addition, many systems do not have the
necessary combination of strobe control and image analysis for full drop-in-flight evaluation.
This paper includes a discussion of an integrated machine-vision based system for visualization and measurement of
drops-in-flight that can be used with any frequency-based jetting system. The strobe is linked to the firing frequency of
the print head, so while it is synchronized, it is independent of the specific print head being inspected.
The imaging system resolves droplets down to 2 picoliters in volume at the highest zoom level. And an open architecture
software package allows for image collection and archiving as well as powerful and flexible image analysis.
This paper will give an overview of the details of this system as well as show some of the system capabilities through
several examples of drop-in-flight analysis.
Image on paper registration measurement and analysis: determining subsystem contributions from a system level measurement
Author(s):
Rakesh Kulkarni;
Abu Islam;
Dan Costanza
Show Abstract
An important print quality attribute of digital printing equipment deals with the absolute position of the printed image
relative to the page. Historically, the most precise method of measuring image to paper (IOP) registration is by scanning
a printed sheet on a flatbed scanner. These measurements have been limited to sheets smaller than the full capacity of
the printer. In addition, the precision of the measurement has been limited by the accuracy of the scanner itself and the
measurement of a few (~4) points on the page have limited the information that can be gathered. The new method
proposed in this paper measures IOP registration throughout the sheet in a more precise manner. In a similar fashion, the
relative position of the image on both the simplex and duplex side of the print can be determined. In addition, the new
method helps link the source of registration errors to individual sub-systems. By generating the individual error sources
from a printed sheet enables the understanding of the percentage contribution of each sub-system, prioritizes efforts to
obtain better IOP performance, finds initial IOP setup errors of a printing engine, compares different technologies
affecting IOP registration in sub-systems and potentially acts as a diagnostic tool for individual sub-systems.
Effect of image path bit depth on image quality
Author(s):
Edgar Bernal;
Robert P. Loce
Show Abstract
Digital Tone Reproduction Curves (TRCs) are applied to digital images for a variety of purposes including compensation
for temporal engine drift, engine-to-engine color balancing, user preference, spatial nonuniformity, and gray balance.
The introduction of one or more compensating TRCs can give rise to different types of image quality defects: Tonal
errors occur when the printed value differs from the intended value; contours occur when the output step size is larger
than the intended step size; pauses occur when two adjacent gray levels map to the same output level. Multiple-stage
TRCs are implemented when compensation operations are performed independently, such as independent adjustment for
temporal variation and user preference. Multiple TRCs are often implemented as independent operations to avoid
complexity within an image path. The effect of each TRC cascades as an image passes through the image path. While
the original image possesses given and assumed desirable quantization properties, the image passed through cascaded
TRCs can possess tonal errors and gray level step sizes associated with a much lower bit-depth system. In the present
study, we quantify errors (tonal errors and changes in gray-level step size) incurred by image paths with cascaded TRCs.
We evaluate image paths at various bit depths. We consider real-life scenarios in which the local gray-level slope of
cascaded compensating TRCs can implement an increase by as much as 200% and decrease by as much as 66%.
Determination of optimal coring values from psychophysical experiments
Author(s):
Hyung Jun Park;
Zygmunt Pizlo;
Jan P. Allebach
Show Abstract
The use of color electrophotographic (EP) laser printing systems is growing because of their declining cost.
Thus, the print quality of color EP laser printers is more important than ever before. Since text and lines are
indispensable to print quality, many studies have proposed methods for measuring these print quality attributes.
Toner scatter caused by toner overdevelopment in color EP laser printers can significantly impact print quality.
A conventional approach to reduce toner overdevelopment is to restrict the color gamut of printers. However,
this can result in undesired color shifts and the introduction of halftone texture in light regions. Coring, defined
as a process whereby the colorant level is reduced in the interior of text or characters, is a remedy for these
shortcomings. The desired amount of reduction for coring depends on line width and overall nominal colorant
level. In previous work, these amounts were chosen on the basis of data on the perception of edge blur that was
published over 25 years ago.
Detection of worms in error diffusion halftoning
Author(s):
Marius Pedersen;
Fritz Albregtsen;
Jon Yngve Hardeberg
Show Abstract
Digital halftoning is used to reproduce a continuous tone image with a printer. One of these halftoning algorithms,
error diffusion, suffers from certain artifacts. One of these artifacts is commonly denoted as worms. We propose
a simple measure for detection of worm artifacts. The proposed measure is evaluated by a psychophysical
experiment, where 4 images were reproduced using 5 different error diffusion algorithms. The results indicate a
high correlation between the predicted worms and perceived worms.
Characterization of '2D noise' print defect
Author(s):
Ki-Youn Lee;
Yousun Bang;
Heui-Keun Choh
Show Abstract
Graininess and mottle described by ISO 13660 standard are two image quality attributes which are widely used to
evaluate area uniformity in digital prints. In an engineering aspect, it is convenient to classify and analyze high frequency
noise and low frequency noise separately. However, it is continuously reported in previous literature that the ISO
methods do not properly correlate with our perception. Since area quality is evaluated by observing all the characteristics
with a wide range of spectral frequencies in a printed page, it is almost impossible to differentiate between graininess and
mottle separately in our percept.
In this paper, we characterize '2D noise' print defect based on psychophysical experiments which appear as two
dimensional aperiodic fluctuations in digital prints. For each channel of cyan, magenta, and black, our approach is to use
two steps of hybrid filtering to remove invisible image components in the printed area. '2D noise' is computed as the
weighted sum of the graininess and mottle, which two weighting factors are determined by subjective evaluation
experiment. By conducting psychophysical validation experiments, the strong correlation is obtained between the
proposed metric and the perceived scales. The correlation coefficients r2 are 0.90, 0.86, and 0.78 for cyan, magenta and
black, respectively.
Measurement of printer MTFs
Author(s):
Albrecht J. Lindner;
Nicolas Bonnier;
Christophe Leynadier;
Francis Schmitt
Show Abstract
In this paper we compare three existing methods to measure the Modulation Transfer Function (MTF) of a
printing system. Although all three methods use very distinct approaches, the MTF values computed for two of
these methods strongly agree, lending credibility to these methods. Additionally, we propose an improvement to
one of these two methods, initially proposed by Jang & Allebach. We demonstrate that our proposed modification
improves the measurement precision and simplicity of implementation. Finally we discuss the pros and cons of
the methods depending on the intended usage of the MTF.
Image quality assessment by preprocessing and full reference model combination
Author(s):
S. Bianco;
G. Ciocca;
F. Marini;
R. Schettini
Show Abstract
This paper focuses on full-reference image quality assessment and presents different computational strategies aimed to
improve the robustness and accuracy of some well known and widely used state of the art models, namely the Structural
Similarity approach (SSIM) by Wang and Bovik and the S-CIELAB spatial-color model by Zhang and Wandell. We
investigate the hypothesis that combining error images with a visual attention model could allow a better fit of the
psycho-visual data of the LIVE Image Quality assessment Database Release 2. We show that the proposed quality
assessment metric better correlates with the experimental data.
Image quality assessment with manifold and machine learning
Author(s):
Christophe Charrier;
Gilles Lebrun;
Olivier Lezoray
Show Abstract
A crucial step in image compression is the evaluation of its
performance, and more precisely the available way to measure the
final quality of the compressed image. In this paper, a machine
learning expert, providing a final class number is designed. The quality measure is based on
a learned classification process in order to respect the one of
human observers. Instead of computing a final note, our method
classifies the quality using the quality scale recommended by the
UIT. This quality scale contains 5 ranks ordered from 1 (the worst
quality) to 5 (the best quality). This was
done constructing a vector containing many visual attributes.
Finally, the final features vector contains more than 40 attibutes.
Unfortunatley, no study about the existing interactions between the
used visual attributes has been done. A feature selection algorithm
could be interesting but the selection is highly related to the
further used classifier. Therefore, we prefer to perform
dimensionality reduction instead of feature selection. Manifold
Learning methods are used to provide a low-dimensional new
representation from the initial high dimensional feature space.
The classification process is performed on this new low-dimensional
representation of the images. Obtained results are compared to the one
obtained without applying the dimension reduction process to judge the
efficiency of the method.
Three-component weighted structural similarity index
Author(s):
Chaofeng Li;
Alan C. Bovik
Show Abstract
The assessment of image quality is very important for numerous image processing applications, where the goal of
image quality assessment (IQA) algorithms is to automatically assess the quality of images in a manner that is consistent
with human visual judgment. Two prominent examples, the Structural Similarity Image Metric (SSIM) and Multi-scale
Structural Similarity (MS-SSIM) operate under the assumption that human visual perception is highly adapted for
extracting structural information from a scene. Results in large human studies have shown that these quality indices
perform very well relative to other methods. However, the performance of SSIM and other IQA algorithms are less
effective when used to rate amongst blurred and noisy images. We address this defect by considering a three-component
image model, leading to the development of modified versions of SSIM and MS-SSIM, which we call three component
SSIM (3-SSIM) and three component MS-SSIM (3-MS-SSIM).
A three-component image model was proposed by Ran and Farvardin, [13] wherein an image was decomposed into
edges, textures and smooth regions. Different image regions have different importance for vision perception, thus, we
apply different weights to the SSIM scores according to the region where it is calculated. Thus, four steps are executed:
(1) Calculate the SSIM (or MS-SSIM) map. (2) Segment the original (reference) image into three categories of regions
(edges, textures and smooth regions). Edge regions are found where a gradient magnitude estimate is large, while smooth
regions are determined where the gradient magnitude estimate is small. Textured regions are taken to fall between these
two thresholds. (3) Apply non-uniform weights to the SSIM (or MS-SSIM) values over the three regions. The weight for
edge regions was fixed at 0.5, for textured regions it was fixed at 0.25, and at 0.25 for smooth regions. (4) Pool the
weighted SSIM (or MS-SSIM) values, typically by taking their weighted average, thus defining a single quality index for
the image (3-SSIM or 3-MS-SSIM).
Our experimental results show that 3-SSIM (or 3-MS-SSIM) provide results consistent with human subjectivity
when finding the quality of blurred and noisy images, and also deliver better performance than SSIM (and MS-SSIM) on
five types of distorted images from the LIVE Image Quality Assessment Database.
An image similarity metric based on quadtree homogeneity analysis
Author(s):
Eric P. Lam;
Thai N Luong;
Mark P. Miller;
Francis Tom
Show Abstract
Comparing two similar images is often needed to evaluate the effectiveness of an image processing algorithm. But,
there is no one widely used objective measure. In many papers, the mean squared error (MSE) or peak signal to noise
ratio (PSNR) are used. These measures rely entirely on pixel intensities. Though these measures are well understood
and easy to implement, they do not correlate well with perceived image quality. This paper will present an image quality
metric that analyzes image structure rather than entirely on pixels. It extracts image structure with the use of a recursive
quadtree decomposition. A similarity comparison function based on contrast, luminance, and structure will be presented.
Most apparent distortion: a dual strategy for full-reference image quality assessment
Author(s):
Eric C. Larson;
Damon M. Chandler
Show Abstract
The mainstream approach to image quality assessment has centered around accurately modeling the single most relevant
strategy employed by the human visual system (HVS) when judging image quality (e.g., detecting visible differences;
extracting image structure/information). In this paper, we suggest that a single strategy may not be sufficient; rather,
we advocate that the HVS uses multiple strategies to determine image quality. For images containing near-threshold
distortions, the image is most apparent, and thus the HVS attempts to look past the image and look for the distortions (a
detection-based strategy). For images containing clearly visible distortions, the distortions are most apparent, and thus
the HVS attempts to look past the distortion and look for the image's subject matter (an appearance-based strategy).
Here, we present a quality assessment method (MAD: Most Apparent Distortion) which attempts to explicitly model
these two separate strategies. Local luminance and contrast masking are used to estimate detection-based perceived
distortion in high-quality images, whereas changes in the local statistics of spatial-frequency components are used to
estimate appearance-based perceived distortion in low-quality images. We show that a combination of these two measures
can perform well in predicting subjective ratings of image quality.
Low level features for image appeal measurement
Author(s):
Pere Obrador;
Nathan Moroney
Show Abstract
Image appeal may be defined as the interest that a photograph generates when viewed by human observers,
incorporating subjective factors on top of the traditional objective quality measures. User studies were conducted in
order to identify the right features to use in an image appeal measure; these studies also revealed that a photograph may
be appealing even if only a region/area of the photograph is actually appealing. Due to the importance of faces regarding
image appeal, a detailed study of a set of face features is also presented, including face size, color and smile detection.
Extensive experimentation helped identify a good set of low level features, which are described in depth. These features
were optimized using extensive ground truth generated from sets of consumer photos covering all possible appeal levels,
by observers with a range of expertise in photography.
SCID: full reference spatial color image quality metric
Author(s):
S. Ouni;
M. Chambah;
M. Herbin;
E. Zagrouba
Show Abstract
The most used full reference image quality assessments are error-based methods. Thus, these measures are performed by
pixel based difference metrics like Delta E ( E), MSE, PSNR, etc. Therefore, a local fidelity of the color is defined.
However, these metrics does not correlate well with the perceived image quality. Indeed, they omit the properties of the
HVS. Thus, they cannot be a reliable predictor of the perceived visual quality. All this metrics compute the differences
pixel to pixel. Therefore, a local fidelity of the color is defined. However, the human visual system is rather sensitive to a
global quality.
In this paper, we present a novel full reference color metric that is based on characteristics of the human visual system by
considering the notion of adjacency. This metric called SCID for Spatial Color Image Difference, is more perceptually
correlated than other color differences such as Delta E. The suggested full reference metric is generic and independent of
image distortion type. It can be used in different application such as: compression, restoration, etc.
An evaluation of interactive image matting techniques supported by eye tracking
Author(s):
Christoph Rhemann;
Margrit Gelautz;
Bernhard Fölsner
Show Abstract
Recently, the quantitative evaluation of interactive single image matting techniques has become possible by the
introduction of high-quality ground truth datasets. However, quantitative comparisons conducted in previous
work are based on error metrics (e.g. sum of absolute differences) that are not necessarily correlated to the visual
quality of the image as perceived by the user. This motivates research to better understand the perception of
errors inherent to matting algorithms, in order to provide the ground for a future design of error metrics that
better reflect the subjective impression of the human observer.
In this work we gain novel insights into the perception of errors due to imperfect matting results. To investigate
these errors, we compare two recent state-of-the-art matting algorithms in a user study. We use an eye-tracker
to reveal details of the decision making of the users. The data acquired in the user study show a considerable
correlation between expert knowledge in photography and the ability of the user to detect errors in the image.
This is also reflected in the eye-tracking data which reveals different types of scanning paths dependent on the
experience of the user.
Perception of detail in 3D images
Author(s):
Ingrid Heynderickx;
Ronald Kaptein
Show Abstract
A lot of current 3D displays suffer from the fact that their spatial resolution is lower compared to their 2D
counterparts. One reason for this is that the multiple views needed to generate 3D are often spatially multiplexed.
Besides this, imperfect separation of the left- and right-eye view leads to blurring or ghosting, and therefore to
a decrease in perceived sharpness. However, people watching stereoscopic videos have reported that the 3D
scene contained more details, compared to the 2D scene with identical spatial resolution. This is an interesting
notion, that has never been tested in a systematic and quantitative way. To investigate this effect, we had people
compare the amount of detail ("detailedness") in pairs of 2D and 3D images. A blur filter was applied to one
of the two images, and the blur level was varied using an adaptive staircase procedure. In this way, the blur
threshold for which the 2D and 3D image contained perceptually the same amount of detail could be found.
Our results show that the 3D image needed to be blurred more than the 2D image. This confirms the earlier
qualitative findings that 3D images contain perceptually more details than 2D images with the same spatial
resolution.
Perception of time variable quality of scene objects
Author(s):
Leif A. Ronningen;
Erlend Heiberg
Show Abstract
The paper focuses on testing the user perception of time-variable quality of scene objects. The scenes are generated and
presented by the DMP (Distributed Multimedia Plays) packet-based system for futuristic continental multimedia
collaboration. DMP guarantees maximum user-to-user delay and minimum scene quality. Time-variable scene quality
control is obtained by adapting the composition and object resolution to traffic load in the network, by controlled
dropping of sub-objects in network nodes, and by admission control. The perceived quality of scenes was tested using an
existing DMP performance (simulation) model that generates two-dimensional random distributions for the frequency of
overloads in the network nodes versus packet drop rate and duration. Video clips with time-varying scene quality were
synthesized, and showed to test persons. Four sub-objects of spatial resolution 1400 x 1050 pixels, and temporal
resolution of 60Hz was applied. Sub-objects were dropped in the network, and missing sub-objects were regenerated by
standard linear interpolation techniques. Test persons could perceive a moderate average quality reduction 13 on a 0-100
quality scale when 75% of the sub-objects were dropped and interpolated. To improve the quality, edge detection and
correction was added. The test persons could perceive a small average quality reduction of 8 on a 0-100 quality scale
when 75% of the sub-objects were dropped.
Scanner image quality profiling
Author(s):
Chengwu Cui
Show Abstract
When using a document scanner, scan image quality is often unknown to the end user of the scanned image. Document
scanners may employ different imaging technologies that can result in different image characteristics. Variability of
scanner parts and the manufacturing process may also create variability of the scanned image quality from machine to
machine. Image quality of the same scanner may also change as it ages and becomes contaminated. If the scanned image
is used for human viewing, the resulting image quality variability may not be mission critical other than being a visual
annoyance because the human visual system has superb adaptation and segmentation capability. However, if the scanned
image is used for machine recognition or for printing, the image quality variability may become important and even
mission critical. Here we propose a framework to profile the scanner image quality and tag the scanned image with the
IQ profile. We review the potential quantified aspects of scan image quality and propose a method of characterization
with examples.
Weighting of field heights for sharpness and noisiness
Author(s):
Brian W. Keelan;
Elaine W. Jin
Show Abstract
Weighting of field heights is important in cases when a single numerical value needs to be calculated that characterizes
an attribute's overall impact on perceived image quality. In this paper we report an observer study to derive the
weighting of field heights for sharpness and noisiness.
One-hundred-forty images were selected to represent a typical
consumer photo space distribution. Fifty-three sample points were sampled per image, representing field heights of 0,
14, 32, 42, 51, 58, 71, 76, 86% and 100%. Six observers participated in this study. The field weights derived in this
report include both: the effect of area versus field height (which is a purely objective, geometric factor); and the effect of
the spatial distribution of image content that draws attention to or masks each of these image structure attributes. The
results show that relative to the geometrical area weights, sharpness weights were skewed to lower field heights, because
sharpness-critical subject matter was often positioned relatively near the center of an image. Conversely, because noise
can be masked by signal, noisiness-critical content (such as blue skies, skin tones, walls, etc.) tended to occur farther
from the center of an image, causing the weights to be skewed to higher field heights.
Identification of image attributes that are most affected with changes in displayed image size
Author(s):
Jae Young Park;
Sophie Triantaphillidou;
Ralph E. Jacobson
Show Abstract
This paper describes an investigation of changes in image appearance when images are viewed at different image sizes
on a high-end LCD device. Two digital image capturing devices of different overall image quality were used for
recording identical natural scenes with a variety of pictorial contents. From each capturing device, a total of sixty four
captured scenes, including architecture, nature, portraits, still and moving objects and artworks under various
illumination conditions and recorded noise level were selected. The test set included some images where camera shake
was purposefully introduced. An achromatic version of the image set that contained only lightness information was
obtained by processing the captured images in CIELAB space. Rank order experiments were carried out to determine
which image attribute(s) were most affected when the displayed image size was altered. These evaluations were carried
out for both chromatic and achromatic versions of the stimuli. For the achromatic stimuli, attributes such as contrast,
brightness, sharpness and noisiness were rank-ordered by the observers in terms of the degree of change. The same
attributes, as well as hue and colourfulness, were investigated for the chromatic versions of the stimuli. Results showed
that sharpness and contrast were the two most affected attributes with changes in displayed image size. The ranking of
the remaining attributes varied with image content and illumination conditions. Further, experiments were carried out to
link original scene content to the attributes that changed mostly with changes in image size.
Simulation of film media in motion picture production using a digital still camera
Author(s):
Arne M. Bakke;
Jon Y. Hardeberg;
Steffen Paul
Show Abstract
The introduction of digital intermediate workflow in movie production has made visualization of the final image
on the film set increasingly important. Images that have been color corrected on the set can also serve as a basis
for color grading in the laboratory. In this paper we suggest and evaluate an approach that has been used to
simulate the appearance of different film stocks. The GretagMacbeth Digital ColorChecker was captured using
both a Canon EOS 20D camera as well as an analog camera. The film was scanned using an Arri film
scanner. The images of the color chart were then used to perform a colorimetric characterization of these devices
using models based on polynomial regression. By using the reverse model of the digital camera and the forward
model of the analog film chain, the output of the film scanner was simulated. We also constructed a direct
transformation using regression on the RGB values of the two devices. A different color chart was then used as
a test set to evaluate the accuracy of the transformations, where the indirect model was found to provide the
required performance for our purpose without compromising the flexibility of having an independent profile for
each device.
Method for measuring the objective quality of the TV-out function of mobile handsets
Author(s):
Mikko Nuutinen;
Pirkko Oittinen
Show Abstract
Digital cameras, printers and displays have their own established methods to measure their performance. Different
devices have their own special features and also different metrics and measuring methods. The real meaning of
measuring data is often not learnt until hands-on experience is available. The goal of this study was to describe a
preliminary method and metrics for measuring the objective image quality of the TV-out function of mobile handsets.
The TV-out application was image browsing.
Image quality is often measured in terms of color reproduction, noise and sharpness and these attributes were also
applied in this study. The color reproduction attribute was studied with color depth, hue reproduction and color accuracy
metrics. The noise attribute was studied with the SNR (signal to noise ratio) and chroma noise metrics. The sharpness
attribute was studied with the SFR (spatial frequency response) and contrast modulation metrics. The measuring data
was gathered by using a method which digitized the analog signal of the TV-out device with a frame grabber card.
Based on the results, the quantization accuracy, chroma error and spatial reproduction of the signal were the three
fundamental factors which most strongly affected the performance of the TV-out device. The quantization accuracy of
the device affects the number of tones that can be reproduced in the image. The quantization accuracy also strongly
affects the correctness of hue reproduction. According to the results, the color depth metric was a good indicator of
quantization accuracy. The composite signal of TV-out devices transmits both chroma and luminance information in a
single signal. A change in the luminance value can change the constant chroma value. Based on the results, the chroma
noise metric was a good indicator for measuring this phenomenon. There were differences between the spatial
reproductions of the devices studied. The contrast modulation was a clear metric for measuring these differences. The
signal sharpening of some TV-out devices hindered the interpretation of SFR data.
Applying image quality in cell phone cameras: lens distortion
Author(s):
Donald Baxter;
Sergio R. Goma;
Milivoje Aleksic
Show Abstract
This paper describes the framework used in one of the pilot studies run under the I3A CPIQ initiative to quantify overall
image quality in cell-phone cameras. The framework is based on a multivariate formalism which tries to predict overall
image quality from individual image quality attributes and was validated in a CPIQ pilot program. The pilot study
focuses on image quality distortions introduced in the optical path of a cell-phone camera, which may or may not be
corrected in the image processing path. The assumption is that the captured image used is JPEG compressed and the cellphone
camera is set to 'auto' mode. As the used framework requires that the individual attributes to be relatively
perceptually orthogonal, in the pilot study, the attributes used are lens geometric distortion (LGD) and lateral chromatic
aberrations (LCA). The goal of this paper is to present the framework of this pilot project starting with the definition of
the individual attributes, up to their quantification in JNDs of quality, a requirement of the multivariate formalism,
therefore both objective and subjective evaluations were used. A major distinction in the objective part from the 'DSC
imaging world' is that the LCA/LGD distortions found in cell-phone cameras, rarely exhibit radial behavior, therefore a
radial mapping/modeling cannot be used in this case.
Low light performance of digital cameras
Author(s):
Bror Hultgren;
Dirk Hertel
Show Abstract
Photospace data previously measured on large image sets have shown that a high percentage of camera phone pictures
are taken under low-light conditions. Corresponding image quality measurements linked the lowest quality to these
conditions, and subjective analysis of image quality failure modes identified image blur as the most important
contributor to image quality degradation.
Camera phones without flash have to manage a trade-off when adjusting shutter time to low-light conditions. The shutter
time has to be long enough to avoid extreme underexposures, but not short enough that hand-held picture taking is still
possible without excessive motion blur. There is still a lack of quantitative data on motion blur. Camera phones often do
not record basic operating parameters such as shutter speed in their image metadata, and when recorded, the data are
often inaccurate. We introduce a device and process for tracking camera motion and measuring its Point Spread Function
(PSF). Vision-based metrics are introduced to assess the impact of camera motion on image quality so that the low-light
performance of different cameras can be compared. Statistical distributions of user variability will be discussed.
Color-blotch noise characterization for CMOS cameras
Author(s):
Reza Safaee-Rad;
M. Aleksic
Show Abstract
Color noise in the form of clusters of color non-uniformity is a major negative quality factor in color images. This type
of noise is significantly more pronounced in CMOS cameras with increasingly smaller pixel sizes (e.g., 1.75μm and
1.4μm pixel sizes). This paper identifies and quantifies temporal noise as the main factor for this type of noise. As well,
it is shown how differences in R/G/B responses and as well possible presence of R/G/B-response non-linearity can
exacerbate color-blotch noise. Furthermore, it is shown how run-time averaging can effectively remove this noise (to a
large extent) from a color image-if capture condition permits.
Photo-response non-uniformity error tolerance testing methodology for CMOS imager systems
Author(s):
Brent McCleary;
Antonio Ortega
Show Abstract
An image sensor system-level pixel-to-pixel photo-response non-uniformity (PRNU) error tolerance method is presented
in this paper. A scheme is developed to determine sensor PRNU acceptability and corresponding sensor application
categorization. Many low-cost imaging systems utilize CMOS imagers with integrated on-chip digital logic for
performing image processing and compression. Due to pixel geometry and substrate material variations, the light
sensitivity of pixels will be non-uniform (PRNU). Excessive variation in the sensitivity of pixels is a significant cause of
the screening rejection for these image sensors. The proposed testing methods in this paper use the concept of
acceptable degradation applied to the camera system processed and decoded images of these sensors. The analysis
techniques developed in this paper give an estimation of the impact of the sensor's PRNU on image quality. This
provides the ability to classify the sensors for different applications based upon their PRNU distortion and error rates.
The human perceptual criteria is used in the determination of acceptable sensor PRNU limits. These PRNU thresholds
are a function of the camera system's image processing (including compression) and sensor noise sources. We use a
Monte Carlo simulation solution and a probability model-based simulation solution along with the sensor models to
determine PRNU error rates and significances for a range of sensor operating conditions (e.g., conversion gain settings,
integration times). We develop correlations between industry standard PRNU measurements and final processed and
decoded image quality thresholds. The results presented in this paper show that the proposed PRNU testing method can
reduce the rejection rate of CMOS sensors. Comparisons are presented on the sensor PRNU failure rates using industry
standard testing methods and our proposed methods.
Improved video image by pixel-based learning for super-resolution
Author(s):
Kenji Kamimura;
Norimichi Tsumura;
Toshiya Nakaguchi;
Hideto Motomura;
Yoichi Miyake
Show Abstract
In recent years, the resolution of display devices has been extremely increased. The resolution of video camera
(except very expensive one), however, is quite lower than that of display since it is difficult to achieve high spatial
resolution with specific frame rate (e.g. 30 frames per second) due to the limited bandwidth. The resolution
of image can be increased by interpolation, such as bi-cubic interpolation, but in this method it is known that
the edges of image are blurred. To create plausible high-frequency details in the blurred image, super-resolution
technique has been studied for a long time.
In this paper, we proose a new algorithm for video super-resolution by considering multi-sensor camera
system. The multi-sensor camera can capture two types video sequence as follow; (a) high-resolution with low
frame rate luminance sequence, (b) low-resolution with high frame rate color sequences. The training pairs for
super-resolution are obtained from these two sequences. The relationships between the high- and low-resolution
frames are trained using pixel-based feature named "texton" and stored in the database with their spatial
distribution. The low-resolution sequences are then represented with texton and each texton is substituted by
searching the trained database to create high-resolution features in output sequences.
The experimental results showed that the proposed method can well reproduce both the detail regions and
sharp edges of the scene. It was also shown that the PSNR of the image obtained by proposed method is improved
compared to the image by bi-cubic interpolation method.
Subjective video quality comparison of HDTV monitors
Author(s):
G. Seo;
C. Lim;
S. Lee;
C. Lee
Show Abstract
HDTV broadcasting services have become widely available. Furthermore, in the upcoming IPTV services, HDTV
services are important and quality monitoring becomes an issue, particularly in IPTV services. Consequently, there have
been great efforts to develop video quality measurement methods for HDTV. On the other hand, most HDTV programs
will be watched on digital TV monitors which include LCD and PDP TV monitors. In general, the LCD and PDP TV
monitors have different color characteristics and response times. Furthermore, most commercial TV monitors include
post-processing to improve video quality. In this paper, we compare subjective video quality of some commercial HD
TV monitors to investigate the impact of monitor type on perceptual video quality. We used the ACR method as a
subjective testing method. Experimental results show that the correlation coefficients among the HDTV monitors are
reasonable high. However, for some video sequences and impairments, some differences in subjective scores were
observed.
Constructing a metrics for blur perception with blur discrimination experiments
Author(s):
Chien-Chung Chen;
Kuei-Po Chen;
Chia-Huei Tseng;
Sheng-Tzung Kuo;
Kuei-Neng Wu
Show Abstract
In this study, we measured blur discrimination threshold at different blur levels. We found that the discrimination
threshold first decreased and then increased again as reference edge width blur increased. This dipper shape of the blur
discrimination threshold vs. reference width functions (TvW) functions can be explained by a divisive inhibition model.
The first stage of the model contains a linear operator whose excitation is the inner product of the image and the
sensitivity profile of the operator. The response of the blur discrimination mechanism is the power function of the
excitation of the linear operator divided by the sum of the divisive inhibition and an additive factor. Changing mean
luminance of the edge has little effect on blur discrimination except at very low luminance. When luminance is low, the
blur discrimination was higher at small reference blur than those measured at medium to high luminance. This difference
diminished at large reference blur. Such luminance effect can be explained by a change in the additive factor in the
model. Reducing contrast of the edge shifted the whole TvW function up vertically. This effect can be explained by the
decrease of gain factors in the linear operator. With these results, we constructed a metric for blur perception from the
divisive inhibition we proposed and tested in this study.
Motion blur perception considering anisotropic contrast sensitivity of human visual system
Author(s):
Shinji Nakagawa;
Toshiya Nakaguchi;
Norimichi Tsumura;
Yoichi Miyake
Show Abstract
In this paper, we measured the anisotropic spatio-velocity CSF of human visual system and applied the measured
CSF to evaluate motion blur of the LCD. Many Gabor stimuli with different contrasts, spatial frequencies, scroll
speeds and angles are displayed on to the LCD and observers are asked whether those stimuli can perceive or
not. The thresholds of those stimuli are defined as the contrast that 50% of observers perceive the stimulus.
Based on this assessment, we obtained the contrast sensitivity given as the inverse of the threshold. By using
the measured spatio-velocity CSFs, we evaluated the anisotropic motion blur characteristics of the LCD.
A geometry calibration and visual seamlessness method based on multi-projector tiled display wall
Author(s):
Yahui Liu;
Qingxuan Jia;
Hanxu Sun;
Jie Su;
Jinling Zhang
Show Abstract
Multi-projector virtual environment based on PC cluster has characteristics of low cost, high resolution and widely visual
angle, which has become a research hotspot in Virtual Reality application. Geometric distortion calibration and seamless
splicing is key problems in multi-projector display. The paper does research on geometry calibration method and edge
blending. It proposes an automatic calibration preprocessing algorithm based on a camera, which projects images to the
regions expected in terms of the relation between a plane surface and a curved surface and texture mapping method. In
addition, overlap regions, which bring about intensity imbalance regions, may be adjusted by an edge blending function.
Implementation indicates that the approach can accomplish geometry calibration and edge blending on an annular
screen.
A facial expression image database and norm for Asian population: a preliminary report
Author(s):
Chien-Chung Chen;
Shu-ling Cho;
Katarzyna Horszowska;
Mei-Yen Chen;
Chia-Ching Wu;
Hsueh-Chih Chen;
Yi-Yu Yeh;
Chao-Min Cheng
Show Abstract
We collected 6604 images of 30 models in eight types of facial expression: happiness, anger, sadness, disgust, fear,
surprise, contempt and neutral. Among them, 406 most representative images from 12 models were rated by more than
200 human raters for perceived emotion category and intensity. Such large number of emotion categories, models and
raters is sufficient for most serious expression recognition research both in psychology and in computer science. All the
models and raters are of Asian background. Hence, this database can also be used when the culture background is a
concern. In addition, 43 landmarks each of the 291 rated frontal view images were identified and recorded. This
information should facilitate feature based research of facial expression. Overall, the diversity in images and richness in
information should make our database and norm useful for a wide range of research.