Show all abstracts
View Session
- Front Matter: Volume 8655
- Image Classification and Recognition I
- Image Classification and Recognition II
- Image Filtering and Enhancement
- Image Compression
- Image Analysis I
- Image Analysis II
- Image Denoising
- Parallel Processing in Image Processing Systems
- Image Interpolation, Super-Resolution, and Inpainting I
- Image Interpolation, Super-Resolution, and Inpainting II
- Interactive Paper Session
Front Matter: Volume 8655
Front Matter: Volume 8655
Show abstract
This PDF file contains the front matter associated with SPIE Proceedings Volume 8655, including the Title Page, Copyright information, Table of Contents, Introduction (if any), and Conference Committee listing.
Image Classification and Recognition I
Target re-identification in low-quality camera networks
Show abstract
Person re-identification through a camera network deals with finding a correct link between consecutive observations
of the same target among different cameras in order to choose the most probable correspondence
among a set of possible matches. This task is particularly challenging in presence of low-resolution camera
networks. In this work, a method for people re-identification in a framework of low-resolution camera network
is presented. The proposed approach can be divided in two parts. First, the illumination changes of a target
while crossing the network is analyzed. The color structure is evaluated using a novel color descriptor, the
Color Structure Descriptor, which describes the differences of dominant colors between two regions of interest.
Afterwards, a new pruning system for the links, the Target Color Structure is proposed. Results shows that
the improvements achieved applying Target Color Structure control are up to 4% for the top rank and up to
16% considering the first eleven more similar candidates.
Robust face recognition algorithm for identifition of disaster victims
Show abstract
We present a robust face recognition algorithm for the identification of occluded, injured and mutilated faces
with a limited training set per person. In such cases, the conventional face recognition methods fall short due to
specific aspects in the classification. The proposed algorithm involves recursive Principle Component Analysis
for reconstruction of afiected facial parts, followed by a feature extractor based on Gabor wavelets and uniform
multi-scale Local Binary Patterns. As a classifier, a Radial Basis Neural Network is employed. In terms of
robustness to facial abnormalities, tests show that the proposed algorithm outperforms conventional face recognition
algorithms like, the Eigenfaces approach, Local Binary Patterns and the Gabor magnitude method. To
mimic real-life conditions in which the algorithm would have to operate, specific databases have been constructed
and merged with partial existing databases and jointly compiled. Experiments on these particular databases
show that the proposed algorithm achieves recognition rates beyond 95%.
Image Classification and Recognition II
Improved image copyright protection scheme exploiting visual cryptography in wavelet domain
Show abstract
The last few years have seen a massive increment in the use of the Internet as a channel for sharing and
transmitting data, thus requiring the need for copyright protection schemes able to preserve the ownership of
the data. The idea of embedding a watermark directly in the data is however unacceptable in various fields
of application, due to the intrinsic degradation introduced by non reversible watermarking schemes. Hence
some zero watermarking schemes have been developed. In this work we propose an optimization of a recent
watermarking method based on visual cryptography, by improving results against most commont types of attacks
and achieving a higher perceptual quality of the extracted mark.
HDR image multi-bit watermarking using bilateral-filtering-based masking
Show abstract
The present paper proposes a blind multi-bit watermarking method for High Dynamic Range (HDR) images. The proposed
approach is designed in order to guarantee the watermark imperceptibility both in the HDR marked image and in its Low
Dynamic Range (LDR) counterpart, being thus robust against significant non-linear distortions such as those performed
by tone-mapping operators (TMOs). In order to do so, the wavelet transform of the Just Noticeable Difference (JND)-
scaled space of the original HDR image is employed as embedding domain. Moreover, a visual mask taking into account
specific aspects of the Human Visual System (HVS) is exploited to improve the quality of the resulting watermarked image.
Specifically, bilateral filtering is used to locate information on the detail part of the HDR image, where the watermark
should be preferably embedded. A contrast sensitivity function is also employed to modulate the watermark intensity
in each wavelet decomposition subband according to its scale and orientation. An extensive set of experimental results
testifies the effectiveness of the proposed scheme in embedding multi-bit watermarks into HDR images without affecting
the visual quality of the original image, while being robust against TMOs.
Body-part estimation from Lucas-Kanade tracked Harris points
Show abstract
Skeleton estimation from single-camera grayscale images is generally accomplished using model-based
techniques. Multiple cameras are sometimes used; however, skeletal points extracted from a single subject using
multiple images are usually too sparse to be helpful for localizing body parts. For this project, we use a single viewpoint
without any model-based assumptions to identify a central source of motion, the body, and its associated extremities.
Harris points are tracked using Lucas-Kanade refinement with a weighted kernel found from expectation maximization.
The algorithm tracks key image points and trajectories and re-represents them as complex vectors describing the motion
of a specific body part. Normalized correlation is calculated from these vectors to form a matrix of graph edge weights,
which is subsequently partitioned using a graph-cut algorithm to identify dependent trajectories. The resulting Harris
points are clustered into rigid component centroids using mean shift, and the extremity centroids are connected to their
nearest body centroid to complete the body-part estimation. We collected ground truth labels from seven participants for
body parts that are compared to the clusters given by our algorithm.
Hue processing in tetrachromatic spaces
Alfredo Restrepo
Show abstract
We derive colour spaces of the hue-colourfulness-luminance type, on the basis of a four-dimensional hypercube.
We derive a chromatic 2D hue that goes together with chromatic saturation and luminance, a toroidal hue
that goes together with a toroidal saturation and colourfulness, and also, a 3D tint that goes together with
colourfulness.
Image Filtering and Enhancement
Decomposition of satellite-derived images for the distinction of cloud types' features
Show abstract
Linear filtering methods using convolution techniques are applied in computer vision, to detect spatial discontinuities in
the intensity of luminance of photograph images. These techniques are based on the following principal: a pixel’s
neighborhood contains information about its intensity. The variation of this intensity provides some information about
the distribution and the possible decomposition of the image in various features. This decomposition relies on the relative
position of the pixel (edge or not) on the image. These principals, integrated into remote sensing analyses, are applied in
this study to differentiate cloud morphological features representing cloud types from a thermal image product (the
Cloud top temperatures) derived from polar orbit satellites’ observations. This approach contrast with that of other
techniques commonly used in satellite cloud classification, and based on optical or thermodynamic properties of the
clouds. The interpretation of the distribution of these cloud morphological features, and their frequency is evaluated
against another cloud classification method relying on cloud optical properties. The results show a relatively good match
between the two classifications. Implications of these results, on the estimation of the impact of cloud shapes’ variations
on the recent climate are discussed.
Locally tuned inverse sine nonlinear technique for color image enhancement
Show abstract
In this paper, a novel inverse sine nonlinear transformation based image enhancement technique is proposed to improve
the visual quality of images captured in extreme lighting conditions. This method is adaptive, local and simple. The
proposed technique consists of four main stages namely histogram adjustment, dynamic range compression, contrast
enhancement and nonlinear color restoration. Histogram adjustment on each spectral band is performed to belittle the
effect of illumination. Dynamic range compression is accomplished by an inverse sine nonlinear function with a locally
tunable image dependent parameter based on the local statistics of each pixel’s neighborhood regions of the luminance
image. A nonlinear color restoration process based on the chromatic information and luminance of the original image is
employed. A statistical quantitative evaluation is performed with the state of the art techniques to analyze and compare
the performance of the proposed technique. The proposed technique is also tested on face detection in complex lighting
conditions. The results of this technique on images captured in hazy/foggy weather environment are also presented. The
evaluation results confirm that the proposed method can be applied to surveillance, security applications in complex
lighting environments.
Image Compression
Embedding high dynamic range tone mapping in JPEG compression
Show abstract
A method that integrates tone mapping for high dynamic range (HDR) gray-scale images with JPEG
compression is proposed. The tone mapping operator (TMO) is block-based, and structured so that
the same discrete cosine transform (DCT) that is used for the JPEG compression serves to complete
a major part of the tone-mapping operation. Simulations have been done on high dynamic range
images from the Debevec library. Experimental results show the technique successfully tone maps
and compresses simultaneously; the number of bits per pixel is reduced from 32 to an average of
0.67 by the compression, with an average PSNR of 56.3 dB for the compressed tone-mapped images
compared to images that have been only tone-mapped. The output of the proposed method is an
image that requires only limited storage space, and can be decompressed with a standard JPEG
decoder.
Formulation, analysis, and hardware implementation of chaotic dynamics based algorithm for compression and feature recognition in digital images
Show abstract
In this paper we will discuss the utilization of a set of waveforms derived from chaotic dynamical systems for
compression and feature recognition in digital images. We will also describe the design and testing of an embedded
systems implementation of the algorithm. We will show that a limited set of combined chaotic oscillations are sufficient
to form a basis for the compression of thousands of digital images. We will demonstrate this in the analysis of images
extracted from the solar heliospheric observatory (SOHO), showing that we are able to detect coronal mass ejections
(CMEs) in quadrants of the image data during a severe solar event. We undertake hardware design in order to optimize
the speed of the algorithm, taking advantage of its parallel nature. We compare the calculation speed of the algorithm in
compiled C, enhanced Matlab, Simulink, and in hardware.
Quality constraint and rate-distortion optimization for predictive image coders
Show abstract
Next generations of image and video coding methods should of course be efficient in terms of compression, but also
propose advanced functionalities. Among these functionalities such as scalability, lossy and lossless coding, data
protection, Rate Distortion Optimization (RDO) and Rate Control (RC) are key issues. RDO aims at optimizing
compression performances, while RC mechanism enables to exactly compress at a given rate. A less common
functionality than RC, but certainly more helpful, is Quality Control (QC): the constraint is here given by the quality. In
this paper, we introduce a joint solution for RDO and QC applied to a still image codec called Locally Adaptive
Resolution (LAR), providing scalability both in resolution and SNR and based on a multi-resolution structure. The
technique does not require any additional encoding pass. It relies on a modeling and estimation of the prediction errors
obtained in an early work. First, quality constraint is applied and propagated through the whole resolution levels called
pyramid. Then, the quantization parameters are deduced considering inter and intra pyramid level relationships. Results
show that performances of the proposed method are very close to an exhaustive search solution.
Image Analysis I
Visual quality analysis for images degraded by different types of noise
Show abstract
Modern visual quality metrics take into account different peculiarities of the Human Visual System (HVS). One of them is
described by the Weber-Fechner law and deals with the different sensitivity to distortions in image fragments with
different local mean values (intensity, brightness). We analyze how this property can be incorporated into a metric PSNRHVS-
M. It is shown that some improvement of its performance can be provided. Then, visual quality of color images
corrupted by three types of i.i.d. noise (pure additive, pure multiplicative, and signal dependent, Poisson) is analyzed.
Experiments with a group of observers are carried out for distorted color images created on the basis of TID2008 database.
Several modern HVS-metrics are considered. It is shown that even the best metrics are unable to assess visual quality of
distorted images adequately enough. The reasons for this deal with the observer’s attention to certain objects in the test
images, i.e., with semantic aspects of vision, which are worth taking into account in design of HVS-metrics.
Graph cut and image intensity-based splitting improves nuclei segmentation in high-content screening
Show abstract
Quantification of phenotypes in high-content screening experiments depends on the accuracy of single cell analysis. In such
analysis workflows, cell nuclei segmentation is typically the first step and is followed by cell body segmentation, feature
extraction, and subsequent data analysis workflows. Therefore, it is of utmost importance that the first steps of high-content
analysis are done accurately in order to guarantee correctness of the final analysis results. In this paper, we present a novel
cell nuclei image segmentation framework which exploits robustness of graph cut to obtain initial segmentation for image
intensity-based clump splitting method to deliver the accurate overall segmentation. By using quantitative benchmarks and
qualitative comparison with real images from high-content screening experiments with complicated multinucleate cells, we
show that our method outperforms other state-of-the-art nuclei segmentation methods. Moreover, we provide a modular
and easy-to-use implementation of the method for a widely used platform.
Near real-time skin deformation mapping
Show abstract
A novel in vivo approach is described that provides large area mapping of the mechanical properties of the skin in
human patients. Such information is important in the understanding of skin health, cosmetic surgery[1], aging, and
impacts of sun exposure.
Currently, several methods have been developed to estimate the local biomechanical properties of the skin, including the
use of a physical biopsy of local areas of the skin (in vitro methods) [2, 3, and 4], and also the use of non-invasive
methods (in vivo) [5, 6, and 7]. All such methods examine localized areas of the skin.
Our approach examines the local elastic properties via the generation of field displacement maps of the skin created
using time-sequence imaging [9] with 2D digital imaging correlation (DIC) [10]. In this approach, large areas of the
skin are reviewed rapidly, and skin displacement maps are generated showing the contour maps of skin deformation.
These maps are then used to precisely register skin images for purposes of diagnostic comparison.
This paper reports on our mapping and registration approach, and demonstrates its ability to accurately measure the skin
deformation through a described nulling interpolation process. The result of local translational DIC alignment is
compared using this interpolation process. The effectiveness of the approach is reported in terms of residual RMS, image
entropy measures, and differential segmented regional errors.
Image Analysis II
Approximations to camera sensor noise
Show abstract
Noise is present in all image sensor data. Poisson distribution is said to model the stochastic nature of the photon arrival process, while it is common to approximate readout/thermal noise by additive white Gaussian noise (AWGN). Other sources of signal-dependent noise such as Fano and quantization also contribute to the overall noise profile. Question remains, however, about how best to model the combined sensor noise. Though additive Gaussian noise with signal-dependent noise variance (SD-AWGN) and Poisson corruption are two widely used models to approximate the actual sensor noise distribution, the justification given to these types of models are based on limited evidence. The goal of this paper is to provide a more comprehensive characterization of random noise. We concluded by presenting concrete evidence that Poisson model is a better approximation to real camera model than SD-AWGN. We suggest further modification to Poisson that may improve the noise model.
Object segmentation using graph cuts based edges features
Show abstract
The paper presents a simple graph cuts algorithm based edges features to object segmentation problems. The user gives some scribbles to background and foreground of an image. Gaussian mixture models(GMMs) are built based on the scribbles. The pixel without scribble belongs to the background or the foreground depending on the relative probability of each pixel. The contribution of our paper is to add edges features to GMMs. The approach is applied with images from the Grab cuts segmentation database. The approach is suitable for images with noise and in the foreground and background with similar colors.
A hybrid skull-stripping algorithm based on adaptive balloon snake models
Show abstract
Skull-stripping is one of the most important preprocessing steps in neuroimage analysis. We proposed a hybrid
algorithm based on an adaptive balloon snake model to handle this challenging task. The proposed framework
consists of two stages: first, the fuzzy possibilistic c-means (FPCM) is used for voxel clustering, which provides
a labeled image for the snake contour initialization. In the second stage, the contour is initialized outside the
brain surface based on the FPCM result and evolves under the guidance of the balloon snake model, which drives
the contour with an adaptive inward normal force to capture the boundary of the brain. The similarity indices
indicate that our method outperformed the BSE and BET methods in skull-stripping the MR image volumes in
the IBSR data set. Experimental results show the effectiveness of this new scheme and potential applications in
a wide variety of skull-stripping applications.
Image Denoising
Fast noise variance estimation by principal component analysis
Stanislav Pyatykh,
Lei Zheng,
Jürgen Hesser
Show abstract
Noise variance estimation is required in many image denoising, compression, and segmentation applications. In this work, we propose a fast noise variance estimation algorithm based on principal component analysis of image blocks. First, we rearrange image blocks into vectors and compute the covariance matrix of these vectors. Then, we use Bartlett's test in order to select the covariance matrix eigenvalues, which correspond only to noise. This allows estimating the noise variance as the average of these eigenvalues. Since the maximum possible number of eigenvalues corresponding to noise is utilized, it is enough to process only a small number of image blocks, which allows reduction of the execution time. The blocks to process are selected from image regions with the smallest variance. During our experiments involving seven state of the art methods, the proposed approach was signi_cantly faster than the methods with similar or higher accuracy. Meanwhile, the relative error of our estimator was always less than 15%. We also show that the proposed method can process images without homogeneous areas.
Spatial-temporal noise reduction method optimized for real-time implementation
Show abstract
Image de-noising in the spatial-temporal domain has been a problem studied in-depth in the field of digital image
processing. However complexity of algorithms often leads to high hardware resource usage, or computational
complexity and memory bandwidth issues, making their practical use impossible. In our research we attempt to solve
these issues with an optimized implementation of a practical spatial-temporal de-noising algorithm. Spatial-temporal
filtering was performed in Bayer RAW data space, which allowed us to benefit from predictable sensor noise
characteristics and reduce memory bandwidth requirements. The proposed algorithm efficiently removes different kinds
of noise in a wide range of signal to noise ratios. In our algorithm the local motion compensation is performed in Bayer
RAW data space, while preserving the resolution and effectively improving the signal to noise ratios of moving objects.
The main challenge for the use of spatial-temporal noise reduction algorithms in video applications is the
compromise between the quality of the motion prediction and the complexity of the algorithm and required memory
bandwidth. In photo and video applications it is very important that moving objects should stay sharp, while the noise is
efficiently removed in both the static background and moving objects. Another important use case is the case when
background is also non-static as well as the foreground where objects are also moving.
Taking into account the achievable improvement in PSNR (on the level of the best known noise reduction
techniques, like VBM3D) and low algorithmic complexity, enabling its practical use in commercial video applications,
the results of our research can be very valuable.
Evolution of image regularization with PDEs toward a new anisotropic smoothing based on half kernels
Show abstract
This paper is dedicated to a new anisotropic diffusion approach for image regularization based on a gradient and two diffusion directions obtained from half Gaussian kernels. This approach results in smoothing an image while preserving edges. From an anisotropic edge detector, built of half Gaussian derivative kernels, we introduce a new smoothing method preserving structures which drives the diffusion function of the angle between the two edge directions and the gradient value. Due to the two directions diffusion used in the control function, our diffusion scheme enables to preserve edges and corners, contrary to other anisotropic diffusion methods. Moreover, parameters of the Gaussian kernel can be tuned to be sufficiently thin extracting precisely edges whereas its length allows detecting in contour orientations which leads to a coherent image regularization. Finally, we present some experimental results and discuss about the choice of the different parameters.
Poisson shot noise parameter estimation from a single scanning electron microscopy image
Show abstract
Scanning electron microscopy (SEM) has an extremely low signal-to-noise ratio leading to a high level of shot noise which makes further processing difficult. Unlike often assumed, the noise stems from a Poisson process and is not Gaussian but depends on the signal level.
A method to estimate the noise parameters of individual images should be found.
Using statistical modeling of SEM noise, a robust optimal noise estimation algorithm is derived.
A non-local means noise reduction filter tuned with the estimated noise parameters on average achieves an 18% lower root-mean-square error than the untuned filter on simulated images.
The algorithm is stable and can adapt to varying noise levels.
Parallel Processing in Image Processing Systems
Parallel algorithms for fast subpixel detection in hyperspectral imagery
Show abstract
We present parallel algorithms for fast subpixel detection of targets in hyperspectral imagery produced by our
Hyperspectral Airborne Tactical Instrument (HATI-2500). The HATI-2500 hyperspectral imaging system has a
blue-enhanced visible-near-IR (VNIR) and a full short-wave IR (SWIR) range response from 400 to 2500 nm.
It has an industry-leading spectral resolution that ranges from 6 nm down to 1.5 nm in the VNIR region.
The parallel detection algorithm selected for processing the hyperspectral data cubes is based on the adaptive
coherence/cosine estimator (ACE). The ACE detector is a robust detector that is built upon the theory of generalized
likelihood ratio testing (GLRT) in implementing the matched subspace detector to unknown parameters
such as the noise covariance matrix. Subspace detectors involve projection transformations whose matrices can
be efficiently manipulated through multithreaded massively parallel processors on modern graphics processing
units (GPU). The GPU kernels developed in this work are based on the CUDA computing architecture. We
constrain the detection problem to a model with known target spectral features and unstructured background.
The processing includes the following steps: 1) scale and offset applied to convert the data from digital numbers
to radiance values, 2) update the background inverse covariance estimate in a line-by-line manner, and 3) apply
the ACE detector for each pixel for binary hypothesis testing. As expected, the algorithm is extremely effective
for homogeneous background, such as open desert areas; and less effective in mixed spectral regions, such as
those over urban areas. The processing rate is shown to be faster than the maximum frame rate of the camera
(100 Hz) with a comfortable margin.
Vasculature segmentation using parallel multi-hypothesis template tracking on heterogeneous platforms
Dong Ping Zhang,
Lee Howes
Show abstract
We present a parallel multi-hypothesis template tracking algorithm on heterogeneous platforms using a layered dispatch programming model. The contributions of this work are: an architecture-specific optimised solution for vasculature structure enhancement, an approach to segment the vascular lumen network from volumetric CTA images and a layered dispatch programming model to free the developers from hand-crafting mappings to particularly constrained execution domains on high throughput architecture. This abstraction is demonstrated through a vasculature segmentation application and can also be applied in other real-world applications.
Current GPGPU programming models define a grouping concept which may lead to poorly scoped lo cal/ shared memory regions and an inconvenient approach to projecting complicated iterations spaces. To improve on this situation, we propose a simpler and more flexible programming model that leads to easier computation projections and hence a more convenient mapping of the same algorithm to a wide range of architectures.
We first present an optimised image enhancement solution step- by-step, then solve a separable nonlinear least squares problem using a parallel Levenberg-Marquardt algorithm for template matching, and perform the energy efficiency analysis and performance comparison on a variety of platforms, including multi-core CPUs, discrete GPUs and APUs. We propose and discuss the efficiency of a layered-dispatch programming abstraction for mapping algorithms onto heterogeneous architectures.
IMPAIR: massively parallel deconvolution on the GPU
Show abstract
The IMPAIR software is a high throughput image deconvolution tool for processing large out-of-core datasets
of images, varying from large images with spatially varying PSFs to large numbers of images with spatially
invariant PSFs. IMPAIR implements a parallel version of the tried and tested Richardson-Lucy deconvolution
algorithm regularised via a custom wavelet thresholding library. It exploits the inherently parallel nature of
the convolution operation to achieve quality results on consumer grade hardware: through the NVIDIA Tesla
GPU implementation, the multi-core OpenMP implementation, and the cluster computing MPI implementation
of the software. IMPAIR aims to address the problem of parallel processing in both top-down and bottom-up
approaches: by managing the input data at the image level, and by managing the execution at the instruction
level. These combined techniques will lead to a scalable solution with minimal resource consumption and maximal
load balancing. IMPAIR is being developed as both a stand-alone tool for image processing, and as a library
which can be embedded into non-parallel code to transparently provide parallel high throughput deconvolution.
Image Interpolation, Super-Resolution, and Inpainting I
Fast texture and structure image reconstruction using the perceptual hash
Show abstract
This paper focuses on the fast texture and structure reconstruction of images. The proposed method, applied to images,
consists of several steps. The first one deals with the extracted textural features of the input images based on the
Law’s energy. The pixels around damaged image regions are clustered using these features, that allow to define the
correspondence between pixels from different patches. Second, cubic spline curve is applied to reconstruct a structure
and to connect edges and contours in the damaged area. The choice of the current pixel to be recovered is decided using
the fast marching approach. The Telea method or modifications of the exemplar based method are used after this
depending on the classification of the regions where to-be-restored pixel is located. In modification to quickly find
patches we use the perceptual hash. Such a strategy allows to get some data structure containing the hashes of similar
patches. This enables us to reduce the search procedure to the procedure for "calculations" of the patch. The proposed
method is tested on various samples of images, with different geometrical features and compared with the state-of-the-art
image inpainting methods; the proposed technique is shown to produce better results in reconstruction of missing small
and large objects on test images.
Improved multichannel up-sampling method for reconstruction based super-resolution
Show abstract
In reconstruction based super-resolution, it is an important issue to up-sample and to merge the high-frequency
information contained in low-resolution images efficiently and without artifact. The conventional up-sampling methods,
which used for registering low-resolution data to high-resolution grid, have the difference between them and ideal upsampling
in observation modeling. In this paper, we analyze the difference and propose a new up-sampling and merging
method which is able to incorporate low-resolution images without loss of the high frequency data and minimizes the
artifact caused by data insufficiency. By this method, regions that the registered high frequency data do not cover are
naturally regularized without using complex regularizers. The experimental results show that the choice of up-sampling
significantly affects the quality of resulting images and the proposed up-sampling method gives better results compared
with conventional simple up-sampling method.
Kernel-based image upscaling method with shooting artifact reduction
Show abstract
This paper describes the interpolation algorithm which contains shooting or ringing artifact suppression based on
windowed sinc interpolator. In general, the windowed sinc interpolator can achieve better performance by using wider
window. However, using wide window causes more ripples that produce unwanted defects such as ringing or shooting
artifact. Therefore, shooting reduction technique is proposed in this paper for using wider windows to improve the
performance without shooting artifact. The proposed algorithm can suppress shooting artifact by using median sinc
interpolator and it can be also used as a post processor for many kernel-based interpolation methods. The resulted image
shows that the proposed algorithm can maintain local details and suppress shooting artifact in the image well.
Image Interpolation, Super-Resolution, and Inpainting II
A stochastic approach for non-rigid image registration
Show abstract
This note describes a non-rigid image registration approach that parametrizes the deformation field by an additive composition
of a similarity transformation and a set of Gaussian radial basis functions. The bases’ centers, variances, and
weights are determined with a global optimization approach that is introduced in this work. This approach consists of
simulated annealing with a particle filter based generator function to perform the optimization. Additionally, a local refinement
is performed to capture the remaining misalignment. The deformation is constrained to be physically meaningful
(i.e., invertible). Results on 2D and 3D data sets demonstrate the algorithm’s robustness to large deformations.
Video inpainting using scene model and object tracking
Show abstract
The problem of automatic video restoration and object removal attract the attention of many researchers. In this paper we
present a new framework for video inpainting. We consider the case when a camera motion is approximately parallel to
the plane of image projection. The scene may consist of a stationary background with a moving foreground, both of
which may require inpainting. Moving objects can move differently, but should not to change their size. A framework
presented in this paper contains the following steps: moving objects identification, moving objects tracking and
background/foreground segmentation, inpainting and, finally, a video rendering. Some results on test video sequence
processing are presented.
Fast DCT-based algorithm for signal and image accurate scaling
Show abstract
A new DCT-based algorithm for signal and image scaling by arbitrary factor is presented. The algorithm is virtually free of boundary effects and implements the discrete sinc-interpolation, which preserves the spectral content of the signal, and therefore is free from interpolation errors. Being implemented through the fast FFT-type DCT algorithm, the scaling algorithm has computational complexity of O(log[σN]) operations per output sample, where N and [σN] are number of signal input and output samples, correspondingly.
Interactive Paper Session
Parallel GPGPU stereo matching with an energy-efficient cost function based on normalized cross correlation
Sarala Arunagiri,
Jaime Jaloma
Show abstract
Stereo matching is a heavily investigated topic in computer vision because of its wide range of applicability and its high computational complexity. In this paper we explore an energy-efficient technique that is suitable for parallel (GPGPU) stereo matching of remotely-sensed radar images. A major issue with images captured by remote sensing, e.g., by airborne radars, is that they are likely to have speckle noise, which is undesired information that contaminates and distorts the image. Although there are filters that alleviate the effects of speckle noise, they introduce a certain amount of image distortion. It has been demonstrated that cost functions based on Normalized Cross Correlation (NCC) can be used for accurate stereo matching in the presence of speckle noise, thus, we explored such a function for passive stereo matching. Accordingly, this paper presents a new NCC-based cost function that minimizes the number of floating-point operations utilized for stereo matching and compares its performance to that of the original NCC cost function. This evaluation is achieved via experiments in which these cost functions are employed by GPGPU stereo matching codes that use the simulated annealing algorithm. Performance comparison is conducted in terms of power and energy consumption and execution time. The dataset consisted of image pairs with two sizes. The smaller images were approximately one megapixel each and the larger image pair was over 42 megapixels each. For the smaller image pairs there was a negligible difference between the floating-point and the integer versions of the code in terms of both the execution time and power consumption. However, for the larger image pair with over 42 megapixels, there was a substantial (nearly 70%) improvement in terms of execution time and energy consumption. In addition, we also investigated the influence of the compute capability of the GPGPU on the results above. Two NVIDIA GPGPUs: a Fermi-based GPGPU with compute capability 2.0 and a more power efficient Kepler-based GPGPU, with compute capability 3.0, were used for this purpose. For the approximately one-megapixel image pairs, the Fermi-based Tesla C2075, had a larger execution time (20%), power consumption (22%), and energy consumption (40%) compared to the newer Kepler-based, GeForce GTX 670.
Hyperspectral bands prediction based on inter-band spectral correlation structure
Show abstract
Hyperspectral imaging has been widely studied in many applications; notably in
climate changes, vegetation, and desert studies. However, such kind of imaging brings a huge
amount of data, which requires transmission, processing, and storage resources for both
airborne and spaceborne imaging. Compression of hyperspectral data cubes is an effective
solution for these problems. Lossless compression of the hyperspectral data usually results in
low compression ratio, which may not meet the available resources; on the other hand, lossy
compression may give the desired ratio, but with a significant degradation effect on object
identification performance of the hyperspectral data. Moreover, most hyperspectral data
compression techniques exploits the similarities in spectral dimensions; which requires bands
reordering or regrouping, to make use of the spectral redundancy. In this paper, we analyze
the spectral cross correlation between bands for AVIRIS and Hyperion hyperspectral data;
spectral cross correlation matrix is calculated, assessing the strength of the spectral matrix, we
propose new technique to find highly correlated groups of bands in the hyperspectral data
cube based on "inter band correlation square", and finally, we propose a new technique of
band regrouping based on correlation values weights for different group of bands as network
of correlation.
A GPU based implementation of direct multi-bit search (DMS) screen algorithm
Show abstract
In this paper, we study the feasibility for using programmable Graphics Processing Unit (GPU) technology
for image halftoning, in particular implementing the computationally intense Direct Multi-bit Search (DMS)
Screen algorithm. Multi-bit screening is an extension of binary screening, in which every pixel in continuoustone
image can be rendered to one among multiple output states. For example, a 2 bit printer is capable
of printing with four different drop sizes. In our previous work, we have extended the Direct Binary Search
(DBS) to the multi-bit case using Direct Multi-bit Search (DMS) where at every pixel the algorithm chooses the
best drop output state to create a visually pleasing halftone pattern without any user defined guidance. This
process is repeated throughout the entire range of gray levels while satisfying the stacking constraint to create
a high quality multi-bit screen (dither mask). In this paper, we illustrate how employing Graphics Processing
Units (GPU) can speed-up intensive DMS image processing operations. Particularly, we illustrate how different
modules can be been parallelized. The main goal of many of the previous articles regarding DBS is to decrease
the execution time of the algorithm. One of the most common approaches is to decrease the neighborhood size
or filter size. The proposed parallel approach allows us to use a large neighborhood and filter size, to achieve
the highest halftone quality, while having minimal impact on performance. In addition, we also demonstrate
processing several non-overlapping neighborhoods in parallel, by utilizing the GPU's parallel architecture, to
further improve the computational efficiency.
Ore minerals textural characterization by hyperspectral imaging
Show abstract
The utilization of hyperspectral detection devices, for natural resources mapping/exploitation through remote sensing
techniques, dates back to the early 1970s. From the first devices utilizing a one-dimensional profile spectrometer,
HyperSpectral Imaging (HSI) devices have been developed. Thus, from specific-customized devices, originally
developed by Governmental Agencies (e.g. NASA, specialized research labs, etc.), a lot of HSI based equipment are
today available at commercial level. Parallel to this huge increase of hyperspectral systems development/manufacturing,
addressed to airborne application, a strong increase also occurred in developing HSI based devices for “ground”
utilization that is sensing units able to play inside a laboratory, a processing plant and/or in an open field. Thanks to this
diffusion more and more applications have been developed and tested in this last years also in the materials sectors. Such
an approach, when successful, is quite challenging being usually reliable, robust and characterised by lower costs if
compared with those usually associated to commonly applied analytical off- and/or on-line analytical approaches. In this
paper such an approach is presented with reference to ore minerals characterization. According to the different phases
and stages of ore minerals and products characterization, and starting from the analyses of the detected hyperspectral
firms, it is possible to derive useful information about mineral flow stream properties and their physical-chemical
attributes. This last aspect can be utilized to define innovative process mineralogy strategies and to implement on-line
procedures at processing level. The present study discusses the effects related to the adoption of different hardware
configurations, the utilization of different logics to perform the analysis and the selection of different algorithms
according to the different characterization, inspection and quality control actions to apply.
Colour modification and colour combination in double-cone colour space
Alfredo Restrepo
Show abstract
We derive a formula for the result of the additive mixture of two colours, in double-cone space. We use Naka-
Rushton law to combine the luminance, circular weighted averaging to combine the hue and two rules of thumb
to get the resulting chromatic saturation.
A study of non-diagonal models for image white balance
Show abstract
White balance is an algorithm proposed to mimic the color constancy mechanism of human perception. However, as
shown by its name, current white balance algorithms only promise to correct the color shift of gray tones to correct
positions; for other color values, white balance algorithms process them as gray tones and therefore produce undesired
color biases. To improve the color prediction of white balance algorithms, in this paper, we propose a 3-parameter nondiagonal
model, named as PCA-CLSE, for white balance. Unlike many previous researches which use the von Kries
diagonal model for color prediction, we proposed applying a non-diagonal model for color correction which aimed to
minimize the color biases while keeping the balance of white color. In our method, to reduce the color biases, we
proposed a PCA-based training method to gain extra information for analysis and built a mapping model between
illumination and non-diagonal transformation matrices. While a color-biased image is given, we could estimate the
illumination and dynamically determine the illumination-dependent transformation matrix to correct the color-biased
image. Our evaluation shows that the proposed PCA-CLSE model can efficiently reduce the color biases.
A comparison between space-time video descriptors
Show abstract
The description of space-time patches is a fundamental task in many applications such as video retrieval or classification. Each space-time patch can be described by using a set of orthogonal functions that represent a subspace, for example a sphere or a cylinder, within the patch. In this work, our aim is to investigate the differences between the spherical descriptors and the cylindrical descriptors. In order to compute the descriptors, the 3D spherical and cylindrical Zernike polynomials are employed. This is important because both the functions are based on the same family of polynomials, and only the symmetry is different. Our experimental results show that the cylindrical descriptor outperforms the spherical descriptor. However, the performances of the two descriptors are similar.
Face recognition based on logarithmic local binary patterns
Show abstract
This paper presents a novel approach to the problem of face recognition that combines the classical Local Binary Pattern (LBP) feature descriptors with image processing in the logarithmic domain and the human visual system. Particularly, we have introduced parameterized logarithmic image processing (PLIP) operators based LBP feature extractor. We also use the human visual system based image decomposition, which is based on the Weber's law to extract features from the decomposed images and combine those with the features extracted from the original images thereby enriching the feature vector set and obtaining improved rates of recognition. Comparisons with other methods are also presented. Extensive experiments clearly show the superiority of the proposed scheme over LBP feature descriptors. Recognition rates as high as 99% can be achieved as compared to the recognition rate of 96.5% achieved by the classical LBP using the AT&T Laboratories face database.
Hyperspectral images lossless compression using the 3D binary EZW algorithm
Show abstract
This paper presents a transform based lossless compression for hyperspectral images which is inspired by Shapiro
(1993)’s EZW algorithm. The proposed compression method uses a hybrid transform which includes an integer
Karhunrn-Loeve transform (KLT) and integer discrete wavelet transform (DWT). The integer KLT is employed to
eliminate the presence of correlations among the bands of the hyperspectral image. The integer 2D discrete wavelet
transform (DWT) is applied to eliminate the correlations in the spatial dimensions and produce wavelet coefficients.
These coefficients are then coded by a proposed binary EZW algorithm. The binary EZW eliminates the subordinate pass
of conventional EZW by coding residual values, and produces binary sequences. The binary EZW algorithm combines
the merits of well-known EZW and SPIHT algorithms, and it is computationally simpler for lossless compression. The
proposed method was applied to AVIRIS images and compared to other state-of-the-art image compression techniques.
The results show that the proposed lossless image compression is more efficient and it also has higher compression ratio
than other algorithms.
A new set of wavelet- and fractals-based features for Gleason grading of prostate cancer histopathology images
Show abstract
Prostate cancer detection and staging is an important step towards patient treatment selection. Advancements in digital
pathology allow the application of new quantitative image analysis algorithms for computer-assisted diagnosis (CAD) on
digitized histopathology images. In this paper, we introduce a new set of features to automatically grade pathological
images using the well-known Gleason grading system. The goal of this study is to classify biopsy images belonging to
Gleason patterns 3, 4, and 5 by using a combination of wavelet and fractal features. For image classification we use
pairwise coupling Support Vector Machine (SVM) classifiers. The accuracy of the system, which is close to 97%, is
estimated through three different cross-validation schemes. The proposed system offers the potential for automating
classification of histological images and supporting prostate cancer diagnosis.
Method and architecture for quantification of bone structure using microscopic image slices
Show abstract
Tills paper presents a new system that reconstructs, visualizes and classifies trabecular bone structure by using microscopic image slices. In this study, we evaluated the structure of a trabecular bone using 3D X-ray imaging after passing through the special image enhancement and de-nosing algorithms. We propose a new simple imaging technique tool for the quantification of structural changes within the micro architecture of human bones by enhancing the characteristics attributes of the bone architecture from μ-CT scans. Computer simulation illustrates that the presented imaging technique has the potential to become a powerful tool to investigate the structure of trabeculae during in vivo measurements.