Show all abstracts
View Session
- Front Matter
- Image Filtering and Enhancement
- Image Analysis
- Image Segmentation and Classification
- Image Transforms and Applications
- Image Processing Systems
- Image Interpolation and Reconstruction
- Image Representation
- Interactive Paper Session
Front Matter
Front Matter: Volume 7870
Show abstract
This PDF file contains the front matter associated with SPIE Proceedings Volume 7870, including the Title Page, Copyright information, Table of Contents, and the Conference Committee listing.
Image Filtering and Enhancement
Video denoising using separable 4D nonlocal spatiotemporal transforms
Show abstract
We propose a powerful video denoising algorithm that exploits temporal and spatial redundancy characterizing
natural video sequences. The algorithm implements the paradigm of nonlocal grouping and collaborative filtering,
where a higher-dimensional transform-domain representation is leveraged to enforce sparsity and thus regularize
the data. The proposed algorithm exploits the mutual similarity between 3-D spatiotemporal volumes constructed
by tracking blocks along trajectories defined by the motion vectors. Mutually similar volumes are grouped
together by stacking them along an additional fourth dimension, thus producing a 4-D structure, termed group,
where different types of data correlation exist along the different dimensions: local correlation along the two
dimensions of the blocks, temporal correlation along the motion trajectories, and nonlocal spatial correlation
(i.e. self-similarity) along the fourth dimension. Collaborative filtering is realized by transforming each group
through a decorrelating 4-D separable transform and then by shrinkage and inverse transformation. In this way,
collaborative filtering provides estimates for each volume stacked in the group, which are then returned and
adaptively aggregated to their original position in the video. Experimental results demonstrate the effectiveness
of the proposed procedure which outperforms the state of the art.
Intelligent edge enhancement using multilayer neural network based on multi-valued neurons
Show abstract
In this paper, we solve the edge enhancement problem using an intelligent approach. We use a multilayer neural network
based on multi-valued neurons (MLMVN) as an intelligent edge enhancer. The problem of neural edge enhancement
using a classical multilayer feedforward neural network (MLF) was already considered by some authors. Since MLMVN
significantly outperforms MLF in terms of learning speed, number of parameters employed, and generalization
capability, it is very attractive to apply it for solving the edge enhancement problem.
The main result which is presented in the paper, is the proven ability of MLMVN to enhance edges corresponding
to a certain edge detection operator. Moreover, it is possible to enhance edges on noisy images ignoring a noisy texture.
It is shown that to learn any edge detection operator using MLMVN, only a single image is required for learning
purposes.
The most important conclusion is that a neural network can learn different edge detection operators from a single
example and then process those images that did not participate in the learning process detecting edges specifically
corresponding to the learned operator with a high accuracy.
Error minimizing algorithms for nearest neighbor classifiers
Show abstract
Stack Filters define a large class of discrete nonlinear filter first introduced in image and signal processing for noise
removal. In recent years we have suggested their application to classification problems, and investigated their
relationship to other types of discrete classifiers such as Decision Trees. In this paper we focus on a continuous
domain version of Stack Filter Classifiers which we call Ordered Hypothesis Machines (OHM), and investigate
their relationship to Nearest Neighbor classifiers. We show that OHM classifiers provide a novel framework in
which to train Nearest Neighbor type classifiers by minimizing empirical error based loss functions. We use
the framework to investigate a new cost sensitive loss function that allows us to train a Nearest Neighbor type
classifier for low false alarm rate applications. We report results on both synthetic data and real-world image
data.
Signal filtering of daily cloud types' trends as derived from satellite images
Show abstract
The relationship between the intensity functions of contiguous pixels of an image is used on daily global clouds satellite
data to extract local edge gradients for cloud types' classification. The images are cloud top temperatures (CTT) derived
from the National Oceanic and Atmospheric Administration/Advanced Very-High-Resolution Radiometer (NOAA-AVHRR)
satellite observations. The cloud type classification method used is a histogram-based gradient scheme
described as the occurrence of low, mid or high edge gradients in a block of pixels. The distribution of these cloud types
is analyzed, then, the consistency of the monthly variations of the cloud type amount estimation is evaluated. A clear
dependence of the cloud type amount signal on the solar zenith angle is noticeable. This dependence, due to the gradual
satellite drift, is removed through a filtering process using the empirical mode decomposition (EMD) method. The EMD
component, associated with the drift or the solar zenith angle change, is filtered out. The cloud types' amount series
corrected show a substantial improvement in their trends.
Image Analysis
Analysing wear in carpets by detecting varying local binary patterns
Show abstract
Currently, carpet companies assess the quality of their products based on their appearance retention capabilities.
For this, carpet samples with different degrees of wear after a traffic exposure simulation process are rated with
wear labels by human experts. Experts compare changes in appearance in the worn samples to samples with
original appearance. This process is subjective and humans can make mistakes up to 10% in rating. In
search of an objective assessment, research using texture analysis has been conducted to automate the process.
Particularly, Local Binary Pattern (LBP) technique combined with a Symmetric adaptation of the Kullback-
Leibler divergence (SKL) are successful for extracting texture features related to the wear labels either from
intensity and range images. We present in this paper a novel extension of the LBP techniques that improves the
representation of the distinct wear labels. The technique consists in detecting those patters that monotonically
change with the wear labels while grouping the others. Computing the SKL from these patters considerably
increases the discrimination between the consecutive groups even for carpet types where other LBP variations
fail. We present results for carpet types representing 72% of the existing references for the EN1471:1996
European standard.
Line and streak detection on polished and textured surfaces using line integrals
Show abstract
In this paper, a framework for detecting lines in a polished or textured substrate is proposed. Modules for image capture,
rectification, enhancement, and line detection are included. If the surface being examined is specular (mirror-like), the
image capture will be restricted, that is, the camera has to be fixed off-axis in the zenith direction. A module for image
rectification and projection is included to overcome this limitation in order to yield an orthographic image. In addition, a
module for image enhancement that includes high-boost is employed to improve the edge sharpness and decrease the
spatial noise in the image. Finally, a line-integral technique is applied to find the confidence vectors that represent the
spatial positions of the lines of interest. The Full-Width at Half-Max (FWHM) approximation is applied to determine the
corresponding lines in a target image. Experimental results show that our technique has an effective performance on
synthetic and real images. Print quality assessment is the main application of the proposed algorithm; however, it can be
used to detect lines/ streak in prints, on substrate or any type of media where lines are visible.
Spatio-temporal analysis and forward modeling of solar polar plumes in white light
Show abstract
The analysis of the data provided by LASCO-C2 coronagraph onboard the SOHO spatial observatory revealed
the fractal characteristics of many outstanding structures of the solar corona, which is the tiny but extended
envelope of plasma wrapping the Sun. A multiscale analysis of recent image sequences has brought a clearer view
of the evolution and the local structure of these features which results from a two steps projection process of the
2D electronic distribution over the Sun polar caps. To get an insight in the volume density distribution over these
caps and their evolution within time, we used the forward modelling approach based on the present knowledge
about the plasma distribution, the physical process of diffusion and the projection geometry on the field of
view. The analysis provides us with the multifractal characterization of the observed phenomena. In the forward
modelling process the goal is to reconstruct the time sequence of 2D electronic distributions slowly evolving over
the Sun polar caps. We used different methodologies: the inverse Fourier transform of 2D+1D (surface and
time) frequency modelling, the evolving multiscale synthesis with Gaussian wavelets and the concealed Markov
approach. Lately a procedure derivate of the Voss generation schema of fBm fractals has been successfully
developed. These different methods are compared and their relative advantages and drawbacks discussed.
Detecting photographic and computer generated composites
Show abstract
Nowadays, sophisticated computer graphics editors lead to a significant increase in the photorealism of images.
Thus, computer generated (CG) images result to be convincing and hard to be distinguished from real ones at
a first glance. Here, we propose an image forensics technique able to automatically detect local forgeries, i.e.,
objects generated via computer graphics software inserted in natural images, and vice versa. We develop a novel
hybrid classifier based on wavelet based features and sophisticated pattern noise statistics. Experimental results
show the effectiveness of the proposed approach.
Imaging using synchrotron radiation for forensic science
Show abstract
Forensic science is already taking benefits from synchrotron radiation (SR) sources in trace evidence analysis.
In this contribution we show a multi-technique approach to study fingerprints from the morphological and
chemical point of view using SR based techniques such as Fourier transform infrared microspectroscopy (FTIRMS),
X-ray fluorescence (XRF), X-ray absorption structure (XAS), and phase contrast microradiography. Both
uncontaminated and gunshot residue contaminated human fingerprints were deposited on lightly doped silicon
wafers and on poly-ethylene-terephthalate foils. For the uncontaminated fingerprints an univariate approach of
functional groups mapping to model FT-IRMS data was used to get the morphology and the organic compounds
map. For the gunshot residue contaminated fingerprints, after a preliminary elemental analysis using XRF,
microradiography just below and above the absorption edge of the elements of interest has been used to map the
contaminants within the fingerprint. Finally, XAS allowed us to determine the chemical state of the different
elements. The next step will be fusing the above information in order to produce an exhaustive and easily
understandable evidence.
Image Segmentation and Classification
PSO-based methods for medical image registration and change assessment of pigmented skin
Steve Kacenjar,
Matthew Zook,
Michael Balint
Show abstract
There are various scientific and technological areas in which it is imperative to rapidly detect and quantify changes in
imagery over time. In fields such as earth remote sensing, aerospace systems, and medical imaging, searching for timedependent,
regional changes across deformable topographies is complicated by varying camera acquisition geometries,
lighting environments, background clutter conditions, and occlusion. Under these constantly-fluctuating conditions, the
use of standard, rigid-body registration approaches often fail to provide sufficient fidelity to overlay image scenes
together. This is problematic because incorrect assessments of the underlying changes of high-level topography can
result in systematic errors in the quantification and classification of interested areas.
For example, in the current naked-eye detection strategies of melanoma, a dermatologist often uses static morphological
attributes to identify suspicious skin lesions for biopsy. This approach does not incorporate temporal changes which
suggest malignant degeneration. By performing the co-registration of time-separated skin imagery, a dermatologist may
more effectively detect and identify early morphological changes in pigmented lesions; enabling the physician to detect
cancers at an earlier stage resulting in decreased morbidity and mortality.
This paper describes an image processing system which will be used to detect changes in the characteristics of skin
lesions over time. The proposed system consists of three main functional elements: 1.) coarse alignment of timesequenced
imagery, 2.) refined alignment of local skin topographies, and 3.) assessment of local changes in lesion size.
During the coarse alignment process, various approaches can be used to obtain a rough alignment, including: 1.) a
manual landmark/intensity-based registration method1, and 2.) several flavors of autonomous optical matched filter
methods2. These procedures result in the rough alignment of a patient's back topography. Since the skin is a deformable
membrane, this process only provides an initial condition for subsequent refinements in aligning the localized
topography of the skin. To achieve a refined enhancement, a Particle Swarm Optimizer (PSO) is used to optimally
determine the local camera models associated with a generalized geometric transform. Here the optimization process is
driven using the minimization of entropy between the multiple time-separated images. Once the camera models are
corrected for local skin deformations, the images are compared using both pixel-based and regional-based methods.
Limits on the detectability of change are established by the fidelity to which the algorithm corrects for local skin
deformation and background alterations. These limits provide essential information in establishing early-warning
thresholds for Melanoma detection.
Key to this work is the development of a PSO alignment algorithm to perform the refined alignment in local skin
topography between the time sequenced imagery (TSI). Test and validation of this alignment process is achieved using a
forward model producing known geometric artifacts in the images and afterwards using a PSO algorithm to demonstrate
the ability to identify and correct for these artifacts. Specifically, the forward model introduces local translational,
rotational, and magnification changes within the image. These geometric modifiers are expected during TSI acquisition
because of logistical issues to precisely align the patient to the image recording geometry and is therefore of paramount importance to any viable image registration system. This paper shows that the PSO alignment algorithm is effective in
autonomously determining and mitigating these geometric modifiers. The degree of efficacy is measured by several
statistically and morphologically based pre-image filtering operations applied to the TSI imagery before applying the
PSO alignment algorithm. These trade studies show that global image threshold binarization provides rapid and superior
convergence characteristics relative to that of morphologically based methods.
Image-based segmentation for characterization and quantitative analysis of the spinal cord injuries by using diffusion patterns
Show abstract
In medical imaging, magnetic resonance imaging sequences are able to provide information of the damaged brain
structure and the neuronal connections. The sequences can be analyzed to form 3D models of the geometry and further
including functional information of the neurons of the specific brain area to develop functional models. Modeling offers
a tool which can be used for the modeling of brain trauma from images of the patients and thus information to tailor the
properties of the transplanted cells. In this paper, we present image-based methods for the analysis of human spinal cord
injuries. In this effort, we use three dimensional diffusion tensor imaging, which is an effective method for analyzing the
response of the water molecules. This way, our idea is to study how the injury affects on the tissues and how this can be
made visible in the imaging.
In this paper, we present here a study of spinal cord analysis to two subjects, one healthy volunteer and one spinal cord
injury patient. We have done segmentations and volumetric analysis for detection of anatomical differences. The
functional differences are analyzed by using diffusion tensor imaging. The obtained results show that this kind of
analysis is capable of finding differences in spinal cords anatomy and function.
Descreening using segmentation-based adaptive filtering
Show abstract
Ordered halftone patterns in the original document interact with the periodic sampling of the scanner, producing
objectionable moir´e patterns. These are exacerbated when the copy is reprinted with an ordered halftone pattern.
A simple, small low-pass filter can be used to descreen the image and to correct the majority of moir´e artifacts.
Unfortunately, low-pass filtering affects detail as well, blurring it even further. Adaptive nonlinear filtering based
on image features such as the magnitude and the direction of image gradient can also be used. However, non
careful tuning of such filters could either cause damage to small details while descreeing the halftone areas,
or result in less descreening while sharpening small details. In this paper, we present a new segmentation-based
descreening technique. Scanned images are segmented into text, images and halftone classes using a
multiresolution classification of edge features. The segmentation results guide a nonlinear, adaptive filter to
favor sharpening or blurring of image pixels belonging to different classes. Our experimental results show the
ability of the non-linear, segmentation driven filter of successfully descreening halftone areas while sharpening
small size text contents.
Image Transforms and Applications
Secure annotation for medical images based on reversible watermarking in the Integer Fibonacci-Haar transform domain
Show abstract
The increasing use of digital image-based applications is resulting in huge databases that are often difficult to
use and prone to misuse and privacy concerns. These issues are especially crucial in medical applications. The
most commonly adopted solution is the encryption of both the image and the patient data in separate files
that are then linked. This practice results to be inefficient since, in order to retrieve patient data or analysis
details, it is necessary to decrypt both files.
In this contribution, an alternative solution for secure medical image annotation is presented. The proposed
framework is based on the joint use of a key-dependent wavelet transform, the Integer Fibonacci-Haar transform,
of a secure cryptographic scheme, and of a reversible watermarking scheme.
The system allows: i) the insertion of the patient data into the encrypted image without requiring the knowledge
of the original image, ii) the encryption of annotated images without causing loss in the embedded information,
and iii) due to the complete reversibility of the process, it allows recovering the original image after the mark
removal. Experimental results show the effectiveness of the proposed scheme.
Multi-seam carving via seamlets
David D. Conger,
Mrityunjay Kumar,
Hayder Radha
Show abstract
Seam carving is a powerful retargeting algorithm for mapping images to arbitrary sizes with arbitrary aspect ratios.
Meanwhile, the seamlet transform has been introduced as an efficient image representation for seam-carving-based
retargeting over heterogeneous multimedia devices with a broad range of display sizes. The original seamlet transform
was developed using Haar filters, and hence it enabled traditional single-seam carving by removing a single seam at a
time in a recursive manner until the desired image size was reached. In this paper, we develop a more efficient approach
for seam carving by enabling multi-seam carving, where at each step of the retargeting algorithm, multiple seams are
carved simultaneously. We achieve multi-seam carving by (a) extending the seamlet transform to allow for larger filters,
and (b) employing local circular convolution in the vicinity of the selected seams. We show that by extending the
seamlet transform we can employ popular filterbanks such as Daubechies' wavelets to achieve efficient multi-seam
carving with visual quality that is comparable to single-seam carving using the Haar transform. Furthermore, with multi-seam
carving, the number of iterations needed to achieve a given target size can be reduced significantly.
A new DCT-based algorithm for numerical reconstruction of electronically recorded holograms
Show abstract
A new universal low computational complexity algorithm for numerical reconstruction of holograms recorded in near
diffraction zone is presented. The algorithm implements digital convolution in DCT domain, which makes it virtually
insensitive to boundary effects. It can be used for reconstruction of holograms for arbitrary ratios of hologram size to the
object-to-hologram distance and wavelength to camera pitch and allows image reconstruction in arbitrary scale.
Image Processing Systems
User discrimination in automotive systems
Show abstract
The recently developed dual-view touch screens, which are announced to be installed in cars in a near future, give rise to
completely new challenges in human-machine interaction. The automotive system should be able to identify if the driver
or the passenger is currently interacting with the touch screen to provide a correct response to the touch. The optical
devices, due to availability, acceptance by the users and multifunctional usage, approved to be the most appropriate
sensing technology for driver/passenger discrimination. In this work the prototypic optical user discrimination system is
implemented in the car simulator and evaluated in the laboratory environment with entirely controlled illumination.
Three tests were done for this research. One of them examined if the near-infrared illumination should be switched on
around the clock, the second one if there is a difference in discrimination performance between day, twilight and night
conditions, and the third one examined how the intensive directional lighting influences the performance of the
implemented user discrimination algorithm. Despite the high error rates, the evaluation results show that very simple
computer vision algorithms are able to solve complicated user discrimination task. The average error rate of 10.42%
(daytime with near-infrared illumination) is a very promising result for optical systems.
Image Interpolation and Reconstruction
Wiener crosses borders: interpolation based on second order models
Alvaro Guevara,
Rudolf Mester
Show abstract
Interpolation of signals (arbitrary dimension, here: 2D images) with missing data points is addressed from a
statistical point of view. We present a general framework for which a Wiener-style MMSE estimator can be
seamlessly adapted to deal with problems such as image interpolation (inpainting), reconstruction from sparse
samples, and image extrapolation. The proposed method gives a precise answer on a) how arbitrary can linear
filters can be applied to initially incomplete signals and b) shows the definite way to extend images beyond theirs
borders such that no size reduction occurs if a linear filter/operator is to be applied to the image.
Image interpolation based on a multi-resolution directional map
Show abstract
This paper describes an interpolation method that takes into account the edge orientation in order to avoid
typical interpolation artifacts (jagging, staircase effects...). It is first based on an edge orientation estimation,
performed in the wavelet domain. The estimation uses the multi-resolution features of wavelets to give an
accurate and non-biased description of the frequency characteristics of the edges, as well as their orientation.
The interpolation is then performed, using the edge orientation estimation, to improve a reference interpolation
(cubic-spline for instance). This improvement is carried out by filtering the edges with a Gaussian kernel along
their direction in order to smooth the contour in the direction parallel to the edge, which avoids disturbing
variations across them (jagging and staircase effects). This technique also keeps the sharpness of the transition
in the direction perpendicular to the contour to avoid blur.
Results are presented on both synthetic and real images, showing the visual impact of the presented method on
the quality of interpolated images. Comparisons are made with the usual cubic-spline interpolation, and with
other edge-directed interpolation techniques to discuss the choices that have been made in our method.
Images reconstruction using modified exemplar based method
Show abstract
This paper describes a novel image reconstruction method based on modified exemplar based technique. This
modification allows to choose sub-optimally image-adaptive form and size of the block in order to find similar patches,
number of which is further increased by rotation of these blocks. We show that the efficiency of image reconstruction
depends on the choice of block size for the exemplar based method. Proposed block size selection adaptivity allows to
obtain a smaller reconstruction error than that of the traditional method as well as other state-of-the-art image inpainting
methods. We demonstrate the performance of a new approach via several examples, showing the effectiveness of the
proposed algorithm in removal of small and large objects on the test images.
Image Representation
A graph, non-tree representation of the topology of a gray scale image
Show abstract
The paper provides a method of graph representation of gray scale images. For binary images, it is generally
recognized that not only connected components must be captured, but also the holes. For gray scale images, there
are two kinds of "connected components" - dark regions surrounded by lighter areas and light regions surrounded
by darker areas. These regions are the lower and upper level sets of the gray level function, respectively. The
proposed method represents the hierarchy of these sets, and the topology of the image, by means of a graph. This
graph contains the well-known inclusion trees, but it is not a tree in general. Two standard topological tools are
used. The first tool is cell decomposition: the image is represented as a combination of pixels as well as edges and
vertices. The second tool is cycles: both the connected components and the holes are captured by circular sequences
of edges.
Colour processing in Runge space
Alfredo Restrepo
Show abstract
We do colour image processing in an RGB-derived spherical space with colour attributes given by hue, colourfulness
(as opposed to grayness and somewhat different from saturation) and lightness; we call it Runge space.
This spherical space is as intuitive as more common spaces of the type hue-saturation-luminance (called here
η,Ε,Λ spaces), yet it avoids the continuity problems of the transformation (R, G,B) → (η,Ε,Λ) that result
from normalizing the saturation by the luminance, or of having a geometrically nonhomogeneous space, when
the saturation is left un-normalized. We give Matlab routines for the conversions between colour spaces RGB
and Runge, and present applications of colour modification in Runge space.
Robust image registration for multiple exposure high dynamic range image synthesis
Susu Yao
Show abstract
Image registration is an important preprocessing technique in high dynamic range (HDR) image synthesis. This paper proposed a robust image registration method for aligning a group of low dynamic range images (LDR) that are captured with different exposure times. Illumination change and photometric distortion between two images would result in inaccurate registration. We propose to transform intensity image data into phase congruency to eliminate the effect of the changes in image brightness and use phase cross correlation in the Fourier transform domain to perform image registration. Considering the presence of non-overlapped regions due to photometric distortion, evolutionary programming is applied to search for the accurate translation parameters so that the accuracy of registration is able to be achieved at a hundredth of a pixel level. The proposed algorithm works well for under and over-exposed image registration. It has been applied to align LDR images for synthesizing high quality HDR images..
Interactive Paper Session
Efficiency analysis of DCT-based filters for color image database
Show abstract
Images formed by different systems are often noisy which makes filtering a typical operation of image pre-processing.
In many research papers, filter performance is analyzed for a limited number of standard test images and noise
variances. Here we use a recently created color image database TID2008 that allows assessing filter efficiency for 25
color images corrupted by noise with different values of variance, both i.i.d. and spatially correlated. Besides, this
image database serves the purpose of evaluating different quality metrics including those able to characterize visual
quality of original and processed images considerably better than conventional MSE and PSNR. The study is carried out
for filters based on discrete cosine transform (DCT) able to suppress both i.i.d. and spatially correlated noise depending
upon a way of threshold setting. It is shown that improvement of PSNR (IPSNR) due to filtering is very close for R, G,
and B components of color images and this improvement depends on image content. IPSNR reaches 9 dB for quite
simple images and it is only about 1 dB for highly textural images if initial PSNR=30 dB. Note that IPSNR is larger if
the original PSNR is smaller. The visual quality metric PSNR-HVS-M is studied as well. The metric PSNR-HVS-M
becomes larger due to filtering but in smaller degree than PSNR does. We demonstrate that it is possible to forecast
whether or not visual quality can be improved due to filtering or to detect in advance highly textural images for which
filtering can be not efficient enough. The provided output MSEs are also compared to potential limits calculated
according to the recently proposed methodology. It is demonstrated that for highly textural images the DCT filtering
with 8x8 full overlapping blocks and hard thresholding provides output MSE close to potential limits. The provided and
limit MSEs differ from each other by about 10%. For simpler images, the provided and limit MSEs can differ by
1.5...2.5 times. Analysis is also carried out for spatially correlated noise. It is shown that efficiency of filtering in this
case is lower.
Color image lossy compression based on blind evaluation and prediction of noise characteristics
Show abstract
The paper deals with JPEG adaptive lossy compression of color images formed by digital cameras. Adaptation to noise
characteristics and blur estimated for each given image is carried out. The dominant factor degrading image quality is
determined in a blind manner. Characteristics of this dominant factor are then estimated. Finally, a scaling factor that
determines quantization steps for default JPEG table is adaptively set (selected). Within this general framework, two
possible strategies are considered. A first one presumes blind estimation for an image after all operations in digital
image processing chain just before compressing a given raster image. A second strategy is based on prediction of noise
and blur parameters from analysis of RAW image under quite general assumptions concerning characteristics
parameters of transformations an image will be subject to at further processing stages. The advantages of both strategies
are discussed. The first strategy provides more accurate estimation and larger benefit in image compression ratio (CR)
compared to super-high quality (SHQ) mode. However, it is more complicated and requires more resources. The second
strategy is simpler but less beneficial. The proposed approaches are tested for quite many real life color images acquired
by digital cameras and shown to provide more than two time increase of average CR compared to SHQ mode without
introducing visible distortions with respect to SHQ compressed images.
Unsupervised automated panorama creation for realistic surveillance scenes through weighted mutual information registration
Show abstract
Automated panorama creation usually requires camera calibration or extensive knowledge of camera locations and
relations to each other. Registration problems are often solved by these same camera parameters or the result of complex
point matching schemes. This paper presents a novel automated panorama creation algorithm by using an affine
transformation search based on maximized mutual information (MMI). MMI techniques are often limited to airborne and
satellite imagery or medical images, but we can show that a simple MMI algorithm very well approximates realistic
scenes of varying depth distortion. This study was performed on stationary color surveillance video cameras and proves
extremely worthwhile in any system with limited or no a priori camera-to-camera parameters. This algorithm is quite
robust on a very large range of strict- to nearly-affine related scenes, and provides a great approximation for the overlap
regions in scenes related by a projective homography. Practical considerations were surprisingly significant in molding
the development of this robust and versatile algorithm.
Ellipse detection using an improved randomized Hough transformation
Show abstract
This paper proposes an ellipse detection algorithm based on the analytical solution to the parameters of ellipse in images.
In the first instance, edge detection is processed, from which line segments are extracted. Then the method of finding the
center coordinates of the ellipse is described based on the property of ellipse by using three points voting at a sense of
randomized Hough Transformation (RHT). Finally, an analytical solution of the other three parameters of the ellipse
(semi-major axis length, semi-minor axis length and the angle between the X-axis and the major axis of the ellipse) are
given via coordinate transformation. Based on this solution, we propose the separated parameter voting scheme for
ellipse center and the other three parameters instead of 5 parameters voting scheme of RHT. The experiments show that
the proposed algorithm performs well in various images.
Detection of motion blur direction based on maxima locations for blind deconvolution
Rachel M. Chong,
Toshihisa Tanaka
Show abstract
The blurs in images closely resemble an ideal point spread function (PSF) model. This similarity can be exploited
in the deconvolution process by learning a model that best fits the estimated PSF. In order to achieve this, a
model is selected from the provided training set and then integrated into the reconstruction cost function. In
this paper, we propose to eliminate the need for a training set and instead use a reference PSF (RPSF) in its
place. This eliminates the need for specifying a training set as well as the dependence on estimated quantities.
Furthermore, it is only dependent on the given degraded image assuming that it is uniformly blurred. The
method is tested with motion blurs in different directions since it is one of the most commonly encountered
problems when using consumer cameras. Using the blur support as a priori knowledge, the results show that the
proposed method is capable of accurately determining the motion direction even in the presence of noise. The
reconstruction of the image is achieved by using a modified cost function that also accounts for the contour of
the estimated PSF. Results show that higher image quality and lower PSF estimation error can be obtained.
EM algorithm-based hyperparameters estimator for Bayesian image denoising using BKF prior
Show abstract
This paper is devoted to a novel hyperparameters estimator for bayesian denoising of images using the Bessel
K Forms prior which we recently developed. More precisely, this approach is based on the EM algorithm.
The simulation results show that this estimator offers good performances and is slightly better compared to
the cumulant-based estimator suggested in. A comparative study is carried to show the effectiveness of our
bayesian denoiser based on EM algorithm compared to other denoisers developed in both classical and bayesian
contexts. Our study has been effected on natural and medical images for gaussian and poisson noise removal.
Color image enhancement algorithm based on logarithmic transform coefficient histogram
Show abstract
This paper presents a new technique for color enhancement based on manipulation of the histogram of logarithmic
transform coefficients. The proposed technique is simple but more effective than some existing techniques in most case.
This method is based on the properties of the histogram of DCT coefficients, also use the fact that the relationship
between stimulus and perception is logarithmic and can afford a marriage between enhancement qualities and
computational efficiency. A human visual system-based quantitative measurement of image contrast improvement is
also used to determine the optimal parameters for the algorithm. A number of experimental results are presented to
illustrate the performance of the proposed algorithm.
Neighbourhood-consensus message passing and its potentials in image processing applications
Show abstract
In this paper, a novel algorithm for inference in Markov Random Fields (MRFs) is presented. Its goal is to
find approximate maximum a posteriori estimates in a simple manner by combining neighbourhood influence of
iterated conditional modes (ICM) and message passing of loopy belief propagation (LBP). We call the proposed
method neighbourhood-consensus message passing because a single joint message is sent from the specified
neighbourhood to the central node. The message, as a function of beliefs, represents the agreement of all
nodes within the neighbourhood regarding the labels of the central node. This way we are able to overcome
the disadvantages of reference algorithms, ICM and LBP. On one hand, more information is propagated in
comparison with ICM, while on the other hand, the huge amount of pairwise interactions is avoided in comparison
with LBP by working with neighbourhoods. The idea is related to the previously developed iterated conditional
expectations algorithm. Here we revisit it and redefine it in a message passing framework in a more general
form. The results on three different benchmarks demonstrate that the proposed technique can perform well both
for binary and multi-label MRFs without any limitations on the model definition. Furthermore, it manifests
improved performance over related techniques either in terms of quality and/or speed.
Alternative method for Hamilton-Jacobi PDEs in image processing
A. Lagoutte,
H. Salat,
C. Vachier
Show abstract
Multiscale signal analysis has been used since the early 1990s as a powerful tool for image processing, notably
in the linear case. However, nonlinear PDEs and associated nonlinear operators have advantages over linear
operators, notably preserving important features such as edges in images. In this paper, we focus on nonlinear
Hamilton-Jacobi PDEs defined with adaptive speeds or, alternatively, on adaptive morphological fiters also called
semi-flat morphological operators. Semi-flat morphology were instroduced by H. Heijmans and studied only in
the case where the speed (or equivalently the filtering parameter) is a decreasing function of the luminance. It
is proposed to extend the definition suggested by H. Heijmans in the case of non decreasing speeds. We also
prove that a central property for defining morphological filters, that is the adjunction property, is preserved
while dealing with our extended definitions. Finally experimental applications are presented on actual images,
including connection of thin lines by semi-flat dilations and image filtering by semi-flat openings.
Spatially adaptive alpha-rooting in BM3D sharpening
Show abstract
The block-matching and 3-D filtering (BM3D) algorithm is currently one of the most powerful and effective image denoising procedures. It exploits a specific nonlocal image modelling through grouping and collaborative filtering. Grouping finds mutually similar 2-D image blocks and stacks them together in 3-D arrays. Collaborative filtering produces individual estimates of all grouped blocks by filtering them jointly, through transform-domain shrinkage of the 3-D arrays (groups).
BM3D can be combined with transform-domain alpha-rooting in order to simultaneously sharpen and denoise the image. Specifically, the thresholded 3-D transform-domain coefficients are modified by taking the alpha-root of their magnitude for some alpha > 1, thus amplifying the differences both within and between the grouped blocks. While one can use a constant (global) alpha throughout the entire image, further performance can be achieved by allowing different degrees of sharpening in different parts of the image, based on content-dependent information.
We propose to vary the value of alpha used for sharpening a group through weighted estimates of the low-frequency, edge, and high-frequency content of the average block in the group. This is shown to be a viable approach for image sharpening, and in particular it can provide an improvement (both visually and in terms of PSNR) over its global non-adaptive counterpart.
Joint distributed source-channel coding for 3D videos
Show abstract
This paper presents a distributed joint source-channel 3D video coding system. Our aim is the design of an
efficient coding scheme for stereoscopic video communication over noisy channels that preserves the perceived
visual quality while guaranteeing a low computational complexity. The drawback in using stereo sequences is
the increased amount of data to be transmitted. Several methods are being used in the literature for encoding
stereoscopic video. A significantly different approach respect to traditional video coding has been represented by
Distributed Video Coding (DVC), which introduces a flexible architecture with the design of low complex video
encoders. In this paper we propose a novel method for joint source-channel coding in a distributed approach.
We choose turbo code for our application and study the new setting of distributed joint source channel coding
of a video. Turbo code allows to send the minimum amount of data while guaranteeing near channel capacity
error correcting performance.
In this contribution, the mathematical framework will be fully detailed and tradeoff among redundancy and
perceived quality and quality of experience will be analyzed with the aid of numerical experiments.
Simulating images captured by superposition lens cameras
Show abstract
As the demand for reduction in the thickness of cameras rises, so too does the interest in thinner lens designs.
One such radical approach toward developing a thin lens is obtained from nature's superposition principle as used in the
eyes of many insects. But generally the images obtained from those lenses are fuzzy, and require reconstruction
algorithms to complete the imaging process. A hurdle to developing such algorithms is that the existing literature does
not provide realistic test images, aside from using commercial ray-tracing software which is costly. A solution for that
problem is presented in this paper. Here a Gabor Super Lens (GSL), which is based on the superposition principle, is
simulated using the public-domain ray-tracing software POV-Ray. The image obtained is of a grating surface as viewed
through an actual GSL, which can be used to test reconstruction algorithms. The large computational time in rendering
such images requires further optimization, and methods to do so are discussed.
Features extraction based on Fisher's information
Show abstract
In this paper a novel scheme for extracting the global features from an image. Usually the features are extracted from the
whole image. In the proposed approach, only the image regions conveying information are considered. The two steps
procedure is based on the Fisher's information evaluation computed by linear combination of Zernike expansion
coefficients. Then, by using the region growing algorithm, only high information rate regions are considered. The
considered features are texture, edges, and color. The performances of the proposed scheme has been evaluated by using
the retrieval rate. Experimental results show an increase in the retrieval rate with respect to use the same features
computed on whole image.
An improved RANSAC algorithm using within-class scatter matrix for fast image stitching
Show abstract
An improved RANSAC algorithm using within-class scatter matrix for fast image stitching is proposed in this paper.
First, features described by SIFT are extracted. Next, the Min-cost K-flow algorithm is used to match SIFT points in
different images. Then, the improved RANSAC algorithm with the within-class scatter matrix is used to divide the
matching feature points into two classes: inliers and outliers. Finally, the homography is computed in the set of inliers.
Experiment results show that the improved algorithm can increase the registration speed by some 20 percent with the
same accuracy and robustness comparing to the original RANSAC algorithm.
Enhanced bleed through removal for scanned document images
Show abstract
Back-to-front interference is a common problem in documents, printed on translucent pages with insufficient opacity and
is referred to as bleed through. The present state-of-art algorithms address bleed through based on entropy, entropic
correlation and discriminator analysis. However, a common drawback of such algorithms is their inefficient
processing of documents that are either sparse in terms of content or have a very dark background. Our proposed
algorithm, based on Otsu's binarization method and pixel level classification addresses these problems. Experiments
indicate that our algorithm performs comparable to state-of-the-art for most of the images and better than state-of-the-art
for the low contrast images.
Wavelet-based asphalt concrete texture grading and classification
Show abstract
In this Paper, we introduce a new method for evaluation, quality control, and automatic grading of texture images
representing different textural classes of Asphalt Concrete (AC). Also, we present a new asphalt concrete texture
grading, wavelet transform, fractal, and Support Vector Machine (SVM) based automatic classification and recognition
system. Experimental results were simulated using different cross-validation techniques and achieved an average
classification accuracy of 91.4.0 % in a set of 150 images belonging to five different texture grades.
Image segmentation refinement by modeling in turning function space
Show abstract
This work proposes a different approach for the use of turning function space to change shapes in accordance with shape
descriptions and consistent with spectral information. The main steps are: (1) segmentation; (2) contour extraction;
(3) turning function space transform; (4) classification; (5) shape analysis; and (6) blob enhancement on image space. In
the analysis of shape the boundary is modified based on both image and model and constraints are imposed to portions of
the turning function. Shape modeling can be done by defining criteria such as linearity, angles and sizes. Results on
synthetic examples are presented.
Integrating ensemble empirical mode decomposition and nonlinear anisotropic diffusion filter for speckle noise reduction in underwater sonar images
Show abstract
In this paper, a speckle noise reduction method is presented. The proposed method is based on a combination of
nonlinear anisotropic diffusion filter and Ensemble Empirical Mode Decomposition (EEMD) technique. It
incorporates the advantages of the two techniques. The experimental results on the images speckled by various
levels of noise show that the proposed method is able to significantly improve the performance of nonlinear
anisotropic diffusion filter. Furthermore, it outperforms several well-known speckle reduction algorithms in terms of
noise removal as well as image features preservation.
Extending JPEG-LS for low-complexity scalable video coding
Show abstract
JPEG-LS, the well-known international standard for lossless and near-lossless image compression, was originally
designed for non-scalable applications. In this paper we propose a scalable modification of JPEG-LS and compare
it with the leading image and video coding standards JPEG2000 and H.264/SVC intra for low-complexity
constraints of some wireless video applications including graphics.