INTEGRATING IMAGING AND VISION
FOR CONTENT-SPECIFIC IMAGE ENHANCEMENT
Gianluigi Ciocca, Claudio Cusano, Francesca Gasparini and Raimondo Schettini
DISCo (Dipartimento di Informatica, Sistemistica e Comunicazione)
Università degli Studi di Milano-Bicocca, Via Bicocca degli Arcimboldi 8, 20126 Milano, Italy
Keywords: Automatic image enhancement, white balancing, contrast enhancement, edge sharpening, image
classification, image annotation, redeye removal, face detection, skin detection.
Abstract: The quality of real-world photographs can often be considerably improved by digital image processing .In
this article we describe our approach, integrating imaging and vision, for content-specific image
enhancement. According to our approach, the overall quality of digital photographs is improved by a
modular, image enhancement procedure driven by the image content. Single processing modules can be
considered as autonomous elements. The modules can be combined to improve the overall quality according
to image and defect categories.
1 INTRODUCTION
The great diffusion of digital cameras and the
widespread use of the internet have produced a mass
of digital images depicting a huge variety of
subjects, generally acquired by non-professional
photographers using unknown imaging systems
under unknown lighting conditions. The quality of
these real-world photographs can often be
considerably improved by digital image processing.
Since interactive processes may prove difficult and
tedious, especially for amateur users, an automatic
image enhancement tool would be most desirable.
There are a number of techniques for image
enhancement, including global and local correction
for color balancing, (Buchsbaum, 1980), (Cardei,
1999), (Barnard, 2002), contrast enhancement
(Tomasi, 1998), (Moroney, 2000) and edge
sharpening (Kashyap, 1994), (Polesel, 2000). Other
techniques merge color and contrast corrections,
such as all the Retinex like algorithms (Land, 1977),
(Rahman, 2004), (Rizzi, 2003), (Meylan, 2004).
Rarely, traditional enhancement algorithms available
in the literature are driven by the content of images
(Naccari, 2005). Our interest is related to the design
of content-aware image enhancement for amateur
digital photographs. The underlying idea is that
global and/or local image classification makes it
possible to set the most appropriate image
enhancement strategy according to the content of the
photograph. To this end, we have pragmatically
designed a modular enhancing procedure integrating
imaging and vision techniques. Each module can be
considered as an autonomous element, related to
color, contrast, sharpness and defect removal. These
modules can be combined in a complete
unsupervised manner to improve the overall quality,
according to image and defect categories. The
proposed method is modular so that each step can be
replaced with a more efficient one in future work,
without changing the main structure. Also, the
method can be improved by simply inserting new
modules. The overall procedure is shown in Figure
1, while the single processing modules are described
in the following Sections. The initial global image
classification makes it possible to further refine the
localization of the color regions requiring different
types of color and sharpness corrections. The
following color, contrast, and edge enhancement
modules may exploit image annotation, together
with further image analysis statistics (in same cases
locally adaptive as well). Red eye removal is the
only specific module we have developed for defect
correction in digital photographs. Other modules
related to different acquisition and/or compression
artifacts are under development. In order to achieve
a more pleasing result, a further processing module
boosts the colors of typical regions such as human
skin, grass, and sky. As a final step we also propose
a self-adaptive image cropping module exploiting
192
Ciocca G., Cusano C., Gasparini F. and Schettini R. (2007).
INTEGRATING IMAGING AND VISION FOR CONTENT-SPECIFIC IMAGE ENHANCEMENT.
In Proceedings of the Second International Conference on Computer Vision Theory and Applications - IU/MTSV, pages 192-199
DOI: 10.5220/0002065201920199
Copyright
c
 SciTePress

both visual and semantic information. This module
can be useful for users looking at photographs on
small displays that require a quick preview of the
relevant image area.
2 IMAGE CLASSIFICATION
Image classification makes it possible to use the
most appropriate image enhancement strategy
according to the content of the photograph. To this
end we developed an automatic classification
strategy (Schettini, 2004) based on the analysis of
low-level features, that is, features that can be
automatically computed without any prior
knowledge of the content of the image.
We considered the classes outdoor, indoor, and
close-up which correspond to typologies of images
that require different enhancement approaches in our
image processing chain. The indoor class includes
photographs of rooms, groups of people, and details
in which the context indicates that the photograph
was taken inside. The outdoor class includes natural
landscapes, buildings, city shots and details in which
the context indicates that the photograph was taken
outside. The close-up class includes portraits and
photos of people and objects in which the context
provides little or no information in regards to where
the photo was taken. Examples of images of these
classes are depicted in
Figure 2.
We adopted a decision forest classifier: an
ensemble of decision trees constructed according to
the CART (Classification And Regression Trees)
methodology. The features we used are related to
color (moments of inertia of the color channels in
the HSV color space, and skin color distribution),
texture and edge (statistics on wavelets
decomposition and on edge and texture
distributions), and composition of the image (in
terms of fragmentation and symmetry). To fully
exploit the fact that trees allow a powerful use of
high dimensionality and conditional information, we
take all the features together and let the training
process perform complexity reduction, and detect
any redundancy. Each decision tree has been trained
on bootstrap replicates of a training set composed of
about 4500 photographs manually annotated with
the correct class. Given an image to classify, the
classification results produced by the single trees are
combined applying the majority vote rule. To further
improve the accuracy of the classifier and to avoid
doubtful decisions, we introduced an ambiguity
rejection option in the classification process: an
image is “rejected” if the confidence on the
classification result is below a tuneable threshold.
Figure 1: Workflow of our modular procedure for content- specific image enhancement.
INTEGRATING IMAGING AND VISION FOR CONTENT-SPECIFIC IMAGE ENHANCEMENT
193

Figure 2: Classification results: examples of outdoor
image (left), indoor image (middle), close-up (right).
3 IMAGE ANNOTATION
3.1 Outdoor Image Annotation
For the detection of semantically meaningful regions
in outdoor photographs, we developed a method
which is capable of automatically segmenting the
images by assigning the regions to seven different
classes: sky, skin, vegetation, snow, water, ground,
and buildings (Cusano, 2005), as depicted in Figure
3. Briefly, the process works as follows: the images
are processed by taking a fixed number of partially
overlapping image subdivisions (tiles) for each pixel
that contain it, each of which is then independently
classified by a multi-class Support Vector Machine
(SVM). The results are used to assign the pixel to
one of the categories. Before submitting a tile to the
classifier we computed a description of it in terms of
low-level features. As feature vectors we used a joint
histogram which combines color distribution with
gradient statistics. For classification, we used a
multi-class SVM, constructed according to the “one
per class” strategy. Seven SVM have been trained to
discriminate between the different classes. The
discriminating functions of the single classifiers are
compared to obtain the output of the combined
classifier.
Buildings
Ground
VegetationSnowSkin
UnknownSky Water
Figure 3: Examples of annotated outdoor images.
a b c d
Figure 4: a: original image; b: segmented skin with a recall
strategy. c: segmented skin with a precision strategy; d:
segmented skin with a trade-off strategy.
3.2 Indoor and Close-ups Image
Annotation: Skin Detection
Many different methods for discriminating between
skin pixels and non-skin pixels are available. The
simplest and most often applied method is to build
an “explicit skin cluster” classifier which expressly
defines the boundaries of the skin cluster in certain
color spaces. The underlying hypothesis of methods
based on explicit skin clustering is that skin pixels
exhibit similar color coordinates in a properly
chosen color space. This type of binary method is
very popular since it is easy to implement and does
not require a training phase. The main difficulty in
achieving high skin recognition rates, and producing
the smallest possible number of false positive pixels,
is that of defining accurate cluster boundaries
through simple, often heuristically chosen, decision
rules. We approached the problem of determining
the boundaries of the skin clusters in multiple color
spaces by applying a genetic algorithm. A good
classifier should have high recall and high precision,
but typically, as recall increases, precision decreases.
Consequently, we adopted a weighed sum of
precision and recall as the fitness of the genetic
algorithm. Keeping in mind that different
applications can have sharply different requirements,
the weighing coefficients can be chosen to offer high
recall or high precision or to satisfy a reasonable
trade-off between these two scores according to
application demands (Gasparini, 2006), as illustrated
in Figure 4. In the following applications addressing
image enhancement, we adopted the boundaries
evaluated for recall oriented strategies.
3.3 Indoor and Close-ups Image
Annotation: Face Detection
Face detection in a single image is a challenging task
because the overall appearance of faces ranges
widely in scale, location, orientation and pose, as
well as in facial expressions and lighting conditions
(Rowley, 1998) and (Yang, 2002). Our objective
therefore was not to determine whether or not there
are any faces, but instead to evaluate the possibility
VISAPP 2007 - International Conference on Computer Vision Theory and Applications
194

of having a facial region. To do so, we have trained
an autoassociative neural network (Gasparini, 2005)
to output a score map reflecting the confidence of
the presence of faces in the input image. It is a three
layer linear network, where each pattern of the
training set is presented to both the input and the
output layers, and the whole network has been
trained by a backpropagation sum square error
criterion, on a training set of more than 150 images,
considering not only face images (frontal faces), but
non-face images as well. The network processes
only the intensity image, so that the results are color
independent. To locate faces of different sizes, the
input image is repeatedly scaled down by a factor of
15%, generating a pyramid of subsampled images.
Figure 5 illustrates this preprocessing and the
application of the neural network with a sample
image. The output is obtained with a feedforward
function, and the root mean square error ε, between
output and input is calculated. The performance of
the network is evaluated analysing the True Positive,
versus the False Positive varying the root mean
square error ε. A score map of the input image is
obtained collecting the likeliness that each single
window in the pyramid contains a facial region
evaluated as 1-FP(ε).
4 AUTOMATIC WHITE
BALANCING
Traditional methods of color balancing do not
discriminate between images with true cast (i.e. a
superimposed dominant color) and those with
predominant colors, and are applied in the same way
to all images. This may result in an undesirable
distortion of the chromatic content with respect to
the original scene. To avoid this problem we
developed a reliable and rapid method for
classifying and removing a color cast in a digital
image. (Gasparini, 2004). A multi-step algorithm
classifies the input images as i) no-cast images; ii)
evident cast images; iii) ambiguous cast images
(images with feeble cast, or for which whether or not
the cast exists is a subjective opinion), iv) images
with a predominant color that must be preserved, v)
unclassifiable images. The whole analysis is
performed by preliminary image statistics for color
distribution in the CIELAB color space. To avoid
the mistaken removal of an intrinsic color, regions
previously identified by image annotation as
probably corresponding to skin, sky, sea or
vegetation, are temporarily removed from the
analyzed image. If an evident or ambiguous cast is
found, a cast remover step, which is a modified
version of the white patch algorithm, is applied.
Since the color correction is calibrated on the type of
the cast, an incorrect choice for the region to be
whitened is less likely, and even ambiguous images
can be processed without color distortion. In Figure
Figure 5: The neural network: the pyramid of the scaled input image, a sampled pixel window, preprocessing, consisting in
histogram equalization of the oval part inside the window, and finally the application of the neural network.
INTEGRATING IMAGING AND VISION FOR CONTENT-SPECIFIC IMAGE ENHANCEMENT
195

6 some examples of images processed by our color
balancing procedure are shown.
5 REDEYE REMOVAL
Figure 7: Top row: original images, bottom row, the effect
of our redeye removal procedure.
The redeye effect is a well known problem in
photography. It is often seen in amateur shots taken
with a built-in flash, but the problem is also well
known to professional photographers. Redeye is the
red reflection of the blood vessels in the retina
caused when a strong and sudden light strikes the
eye. Fixing redeye artifacts digitally became an
important skill with the advent of digital
technologies, which permit users to acquire
digitalized images either directly with a digital
camera or by converting traditional photos from
scanners. Also, the widespread use of small devices
with built-in flashes, including cell phones and
handheld computers, produces a large number of
digital photographs that are potentially affected by
redeye. Currently, many image processing software
applications in the market offer redeye removal
solutions. Most of them are semi-automatic or
manual solutions. The user has to either click on the
redeye or draw a box containing the redeye before
the redeye removal algorithm can find the redeye
pixels and correct them (Benati, 1998), (Patti, 1998),
(Hardeberg, 2002). A typical problem with most of
these algorithms is poor pupil segmentation that
leads to unnatural redeye correction. Even with user
interaction, these algorithms sometimes correct
redeye pixels too aggressively, darkening eyelid
areas, or too conservatively, leaving many redeye
pixels uncorrected. The proposed method
(Gasparini, 2005) is modular so that each step can be
removed and substituted with a more efficient one in
future work, without changing the main structure.
Also, it can be improved by simply inserting new
modules. In our enhancement chain it follows our
color balancing algorithm. This phase not only
facilitates the subsequent steps of processing, but
also improves the overall appearance of the output
image. Like several redeye removal algorithms, the
method we developed looks for redeye within the
most likely face regions. The localization of these
candidate regions is obtained by combining, through
a scoring process, the results of a color-based face
detector based on skin segmentation and the face
detector based on a multi-resolution neural network,
Figure 6: Our automatic color balancing procedure.
Cast Removal
No Cast
Intrinsic
color
Cast Cast
Ambiguous
Cast
Cast detector
VISAPP 2007 - International Conference on Computer Vision Theory and Applications
196

working only on the intensity channel. In the final
phase, red eyes are automatically corrected by
exploiting a novel algorithm that has been designed
to remove the unwanted effects while maintaining
the natural appearance of the processed eyes.
6 LOCAL CONTRAST
ENHANCEMENT
The original dynamic range of a scene is generally
constrained into the smaller dynamic range of the
acquisition system. This makes it difficult to design
a global tone correction that is able to enhance both
shadow and highlight details. Several methods for
adjusting image contrast, (Tomasi, 1998), (Moroney,
2000), (Meylan, 2004), (Rahman, 2004), (Rizzi,
2003), have been developed in the field of image
processing for image enhancement. In general, it is
possible to discriminate between two classes of
contrast corrections: global and local corrections.
With global contrast corrections it is difficult to
accommodate both lowlight and highlight details.
The advantage of local contrast corrections is that it
provides a method to map one input value to many
different output values, depending on the values of
the neighbouring pixels and this allows for
simultaneous shadow and highlight adjustments.
Our contrast enhancement method is based on a
local and image dependent exponential correction
(Capra 2006). The simplest exponential correction,
better known as gamma correction, is common in the
image processing field, and consists in elaborating
the input image through a constant power function.
This correction gives good results for totally
underexposed or overexposed images. However,
when both underexposed and overexposed regions
are simultaneously present in an image, this
correction is not satisfactory. As we are interested in
a local correction, the exponent of the gamma
correction used by our algorithm is not a constant.
Instead, it is chosen as a function that depends on the
point to be corrected, on its neighbouring pixels and
on the global characteristics of the image. This
function is also chosen to be edge preserving to
eliminate halo artifacts. Usually it happens,
especially for low quality images with compression
artefacts, that the noise in the darker zones is
enhanced. To overcome this undesirable loss in the
image quality, a further step of contrast
enhancement was added. This step consists of a
stretching and clipping procedure, and an algorithm
to increase the saturation. An example of this
processing is shown in Figure 8.
Figure 8: Left, original image. Right, final image
processed by our whole contrast enhancement procedure.
7 SELECTIVE EDGE
ENHANCEMENT
Digital images are often corrupted by artifacts due to
noise in the imaging system, digitization, and
compression. Smoothing is a widely used technique
to obtain more visually pleasing images, and several
methods have been proposed in the literature to
reduce edge blurring when smoothing is applied.
Among the edge sharpening techniques, the unsharp
masking approach is widely used to improve the
perceptual quality of an image. Even though unsharp
masking is simple and produces good results in
many applications, its main drawback is that it does
not distinguish between significant and non-
significant high frequencies, such as noise, and thus
all these high frequencies are added with the same
weight. As a result, the algorithm applied to the
original low quality image also enhances noise,
digitization effects and blocking artifacts. We
developed a new approach for selective edge
enhancement (Gasparini, 2005) able to perform
image smoothing, which not only preserves but also
enhances the salient details in images. Our algorithm
is based on the consideration that there is a strong
relationship between biological vision and image
rendering. In particular, the image rendering process
is more successful interpreting the original scene and
applying the appropriate transformations. The key
idea is to process the image locally according to
topographic maps obtained by a neurodynamical
model of visual attention, overcoming the tradeoff
between smoothing and sharpening typical of the
traditional approaches. In fact, only high frequencies
corresponding to regions that are non-significant to
our visual system are smoothed while significant
details are sharpened.
INTEGRATING IMAGING AND VISION FOR CONTENT-SPECIFIC IMAGE ENHANCEMENT
197

8 SELECTIVE COLOR
BOOSTING
In many cases, global modification of the image
colors will not result in correct reproduction. For
some objects whose colors are well known,
preferred color reproduction may be required (Hunt,
1977). Several authors, e.g. (Naccari, 2005), have
suggested selective color correction for objects
having a typical color such as human skin, grass, or
sky in order to achieve a more pleasing result.
Kanamori and Kotera (Kanamori, 1991) in
particular, have suggested a smooth selective change
in hue and saturation color attributes. Although
effective, the method they have developed requires a
great deal of practice: for each image to be
processed, the color set to be changed, and the
degree of change itself, must be specified in
numbers. Taking this as our point of departure, we
have developed a soft color cluster editor which
allows the user to correct or modify the image colors
as they appear, in a simple and effective way, until a
satisfactory reproduction is obtained (Schettini,
1995), (Boldrin, 1999). A soft color cluster is
composed of colors that are similar to a selected
color centroid so that the farther a color lies from the
centroid, the less it will be changed in editing. In
order to effectively define the cluster we have
exploited the best medium for color communication,
sighting, and the fact that computer-driven displays
allow the user to select and view the colors forming
composite images on the screen in real time. Visual
interaction allows the user to select the color
centroids, and to define and edit soft color clusters
without considering their internal representation,
physical qualities, or names.
Different image categories (indoor, outdoor and
close-ups) usually require different color
corrections, therefore different image training sets
have been defined and interactively corrected by a
panel of specialized users. For each image class, we
considered only satisfactory image matches, and all
the colors that have been modified more then a given
threshold are used to train a feed forward neural
network. The implicit mapping coded in the trained
neural networks can be applied to correct the colors
of the processed images.
9 ADAPTIVE IMAGE CROPPING
Some of the efforts that have been put on image
adaptation are related to the ROI coding scheme
introduced in JPEG 2000 (Christopoulos, 2000).
Most of the approaches for adapting images only
focused on compressing the whole image in order to
reduce the data transmitted. Few other methods use
an auto-cropping technique to reduce the size of the
image transmitted (Chen, 2003), (Suh, 2003). These
methods decompose the image into a set of spatial
information elements (saliency regions) which are
then displayed serially to help users’ browsing or
searching through the whole image. These methods
are heavily based on a visual attention model
technique that is used to identify the saliency regions
to be cropped. We designed a self-adaptive image
cropping algorithm exploiting both visual and
semantic information (Ciocca, 2007). Visual
information is obtained by a visual attention model,
while semantic information relates to the
automatically assigned image genre and to the
detection of face and skin regions. The processing
steps of the algorithm are firstly driven by the
classification phase and then further specialized with
respect to the annotated face and skin regions.
Figure 9: Examples of cropping areas selected by our
algorithm.
10 CONCLUSIONS
We have described here our approach, integrating
imaging and vision, for content-specific image
enhancement. The key idea is that the most
appropriate enhancement strategy can be applied if
the photographs are semantically annotated. All of
our image processing methods take into account the
content of the photograph to drive the image
enhancement. We have collected a variety of images
for evaluation purpose. Different algorithms and/or
parameter settings have been quantitatively
evaluated whenever possible, or subjectively
evaluated by pair wise comparison. To this end we
have developed a web based system that makes it
possible the comparison and ranking of different
processing results by different users. Our results
indicates that the proposed solution has several
features in terms of effectiveness, friendliness and
robustness that make it an ideal candidate to be
included within software for the management and
VISAPP 2007 - International Conference on Computer Vision Theory and Applications
198

enhancement of digital photo albums by non expert,
amateur photographers.
REFERENCES
Buchsbaum, G., 1980, A spatial processor model for
object color perception, Journal of Franklin Institute
310, pp 1-26.
Cardei, V., Funt, B., Barnard, K., 1999, White Point
Estimation for Uncalibrated Images, Proceedings of
the IS&T/SID Seventh Color Imaging Conference,
Scottsdale, USA, pp. 97 100.
Barnard, K., Cardei, V., Funt, B., 2002 A Comparison of
Computational Color Constancy Algorithms-Part I:
Methodology and Experiments with Synthesized Data,
IEEE Transactions on Image Processing 11 (9), pp.
972-983.
Tomasi, C., Manduchi, R.,1998, Bilateral filtering for gray
and color images, Proc. IEEE Int. Conf. on Computer
Vision, pp. 836-846.
Moroney, N., 2000, Local colour correction using non-
linear masking, IS&T/SID Eighth Color Imaging
Conference, pp. 108-111.
Land, E., (1997), The Retinex Theory of Color Vision,
Scientific American 237, pp.108-129.
Rahman, Z., Jobson, D., Woodell, G. ,2004, Retinex
processing for automatic image enhancement, Journal
of Electronic Imaging, 13, (1), pp. 100-110.
Rizzi, A., Gatta, C., Marini, D., 2003, A new algorithm for
unsupervised global and local color correction, Pattern
Recognition Letters, 24, pp. 1663-1677.
Meylan L., S¨usstrunk, S., 2004 Bio-inspired image
enhancement for natural color images, in IS&T/SPIE
Electronic Imaging 2004. The Human Vision and
Electronic Imaging IX, 5292, pp. 46-56.
Naccari, F., Battiato, S., Bruna, A., Capra, A. Castorina,
A., 2005, Natural Scene Classification for Color
Enhancement, IEEE Transactions on Consumer
Electronics, 5,(1), pp.234-239.
Kashyap, R.L. 1994, A robust variable length nonlinear
filter for edge enhancement and noise smoothing,
Proc. of the 12th IAPR International Conference on
Signal Processing, 143-145.
Polesel, A., Ramponi, G., Mathews, V.J., 2000, Image
Enhancement Via Adaptive Unsharp Masking, IEEE
Transactions on Image Processing, 9, (3), pp.505-510.
Schettini, R., Brambilla, C., Cusano, C., Ciocca, G., 2004,
Automatic classification of digital photographs based
on decision forests, International Journal of Pattern
Recognition and Artificial Intelligence, 18, (5), pp.
819-845.
Cusano, C., Gasparini, F., Schettini, R., 2005, Image
annotation for adaptive enhancement of uncalibrated
color images, Springer, Lecture Notes in Computer
Sciences, 3736, pp. 216-225.
Yang, M. H, Kriegman, D.J., Ahuja, N., 2002, Detecting
Faces in Images: A Survey, IEEE Trans. on Pattern
Analysis and Machine Intelligence, 24, (1), pp. 34-58.
Rowley, H., Baluja, S., Kanade, T., 1998, Neural
Network-Based Face Detection, IEEE Trans. on
Pattern Analysis and Machine Intelligence, 20, (1),
pp.23-28.
Gasparini, F., Schettini, R., 2004, Color Balancing of
Digital Photos Using Simple Image Statistics, Pattern
Recognition, 37, pp. 1201-1217.
Benati, P., Gray, R., Cosgrove, P., 1998, Automated
detection and correction of eye color defects due to
flash illuminatio, US patent 5,748,764 5 May.
Patti, A., Kostantinides, K., Tretter, D., Lin, Q., 1998,
Automatic Digtal Redeye Reduction, Proceedings of
the IEEE International Conference on Image
Processing, 3, pp 55-59.
Hardeberg, J. Y., 2002, Digital red eye removal’, Journal
of Imaging Science and Technology, 46 (4), pp. 375-
381.
Gasparini, F., Schettini, R., 2005, Automatic redeye
removal for smart enhancement of photos of unknown
origin, Springer, Lecture Notes in Computer Sciences,
3736, pp. 226-233.
Capra, A., Castorina, A., Corchs, S., Gasparini, F.,
Schettini, R., 2006, Dynamic range optimization by
local contrast correction and histogram image analysis,
2006 IEEE International Conference on Consumer
Electronics, Las Vegas, USA, (in print).
Gasparini, F., Corchs, S., Schettini, R., 2005, Adaptive
edge enhancement using a neurodynamical model of
visual attention, In Proc. IEEE International
Conference on Image Processing, 3, pp. 972-975,
Hunt, R. W. G., 1977, Colour reproduction by
photography, Reports on Progress in Physics, 40, pp.
1071-1121.
Schettini, R., Barolo, B., Boldrin, E., 1995, A soft color
cluster, Editor, Image Process. Commun. 1(1) pp. 17-
32.
Kanamori, K., Kotera, H., 1991, A method for selective
color control in perceptual color space, J. Imaging
Technol. 35(5) pp. 307-316.
Boldrin, E., Schettini, R., 1999, Cross-media color
matching using neural networks, Pattern Recognition,
32, pp. 465-476.
Christopoulos C., Skodras A., Ebrahimi T., 2000, The
JPEG2000 still image coding system: an overview,
IEEE Trans Cons Elect. vol. 46, no. 4, pp. 1103–
1127,.
Chen L., Xie X., Fan X., Ma W., Zhang H.J., Zhou H.Q.,
2003, A visual attention model for adapting images on
small displays, Multimedia Systems, 9, pp. 353–364.
Suh B., Ling H., Bederson B.B., Jacobs D.W., 2003,
Automatic Thumbnail Cropping and its Effectiveness,
In Proc. UIST’03, pp. 95-104,.
Ciocca, G., Cusano, C., Gasparini, F., Schettini, R., 2007,
Self Adaptive Image Cropping for Small Displays,
Proc. IEEE Int. Conference on Consumer Electronics
(ICCE), Las Vegas, USA, Jan. 10-14, to be published.
INTEGRATING IMAGING AND VISION FOR CONTENT-SPECIFIC IMAGE ENHANCEMENT
199