1,721,220 research outputs found
Medical image registration and stereo vision using mutual information
No full textImage registration is a fundamental problem that can be found in a diverse range of fields within the research community. It is used in areas such as engineering, science, medicine, robotics, computer vision and image processing, which often require the process of developing a spatial mapping between sets of data. Registration plays a crucial role in the medical imaging field where continual advances in imaging modalities, including MRI, CT and PET, allow the generation of 3D images that explicitly outline detailed in vivo information of not only human anatomy, but also human function.\ud
\ud
\ud
\ud
Mutual Information (MI) is a popular entropy-based similarity measure which has found use in a large number of image registration applications. Stemming from information theory, this measure generally outperforms most other intensity-based measures in multimodal applications as it does not assume the existence of any specific relationship between image intensities. It only assumes a statistical dependence. The basic concept behind any approach using MI is to find a transformation, which when applied to an image, will maximise the MI between two images. This thesis presents research using MI in three major topics encompassed by the computer vision and medical imaging field: rigid image registration, stereo vision, and non-rigid image registration.\ud
\ud
\ud
\ud
In the rigid domain, a novel gradient-based registration algorithm (MIGH) is proposed that uses Parzen windows to estimate image density functions and Gauss-Hermite quadrature to estimate the image entropies. The use of this quadrature technique provides an effective and efficient way of estimating entropy while bypassing the need to draw a second sample of image intensities (a procedure required in previous Parzen-based MI registration approaches). It is possible to achieve identical results with the MIGH algorithm when compared to current state of the art MI-based techniques. These results are achieved using half the previously required sample sizes, thus doubling the statistical power of the registration algorithm. Furthermore, the MIGH technique improves algorithm complexity by up to an order of N, where N represents the number of samples extracted from the images.\ud
\ud
\ud
\ud
In stereo vision, a popular passive method of depth perception, new extensions have been pro- posed in order to increase the robustness of MI-based stereo matching algorithms. Firstly, prior probabilities are incorporated into the MI measure to considerably increase the statistical power of the matching windows. The statistical power, directly related to the number of samples, can become too low when small matching windows are utilised. These priors, which are calculated from the global joint histogram, are tuned to a two level hierarchical approach. A 2D match surface, in which the match score is computed for every possible combination of template and matching windows, is also utilised to enforce left-right consistency and uniqueness constraints. These additions to MI-based stereo matching significantly enhance the algorithms ability to detect correct matches while decreasing computation time and improving the accuracy, particularly when matching across multi-spectra stereo pairs.\ud
\ud
\ud
\ud
MI has also recently found use in the non-rigid domain due to a need to compute multimodal non-rigid transformations. The viscous fluid algorithm is perhaps the best method for re- covering large local mis-registrations between two images. However, this model can only be used on images from the same modality as it assumes similar intensity values between images. Consequently, a hybrid MI-Fluid algorithm is proposed to compute a multimodal non-rigid registration technique. MI is incorporated via the use of a block matching procedure to generate a sparse deformation field which drives the viscous fluid algorithm, This algorithm is also compared to two other popular local registration techniques, namely Gaussian convolution and the thin-plate spline warp, and is shown to produce comparable results. An improved block matching procedure is also proposed whereby a Reversible Jump Markov Chain Monte Carlo (RJMCMC) sampler is used to optimally locate grid points of interest. These grid points have a larger concentration in regions of high information and a lower concentration in regions of small information. Previous methods utilise only a uniform distribution of grid points throughout the image
Registration of three dimensional medical images
No full textImage registration is a fundamental problem that can be found in a diverse range of fields within the research community. It is used in areas such as engineering, science, medicine, robotics, computer vision and image processing, which often require the process of developing a spatial mapping between sets of data. In the field of medical imaging, image registration is required to match images acquired from various imaging modalities. Recent advances in these imaging modalities, including MRI, CTI and PET, now allow the generation of 3D images that explicitly outline detailed in vivo information of not only human anatomy, but also metabolic function.\ud
\ud
The amount of time and effort dedicated to the research of medical image registration is a testimony to the importance and significance that this area holds in the medical field. This has consequently lead to the development of new and fascinating opportunities for areas involving diagnosis and therapy. This includes applications such as surgical planning, image guided surgery and surgery simulation. However, the creation of such opportunities would not have been possible without the enormous advances made in computing technology, which is required in order to facilitate efficient 3D image registration.\ud
\ud
A common task within medical image registration is the fusing of the complimentary and synergistic information provided by the various imaging modalities. This process is known as multimodal registration. Another common task is in the registration of images of the same patient taken at different times and/or in different positions. This process is referred to as mono-modal registration and can be used to track any pathological evolution. Other applications include inter-patient registration and registration of a patient's scan with an anatomical atlas. The latter application is extremely useful for further applications such as the statistical analysis of populations and automatic segmentation.\ud
\ud
In a quest to further understand some of the inherent advantages and disadvantages of image registration algorithms, a literature review was undertaken. A classification of registration algorithms was also presented along with the literature review. This classification scheme is based on certain characteristics that a registration algorithm may exhibit. The categories include the algorithm's dimensionality, nature of the registration algorithm, nature and domain of the transformation, user interaction, optimisation procedure, modalities involved, and the type of subject and objects involved in the registration process.\ud
\ud
Traditional registration methods were based on either manual methods or the use of fiducial markers. These methods either produced a poor accuracy or a greater accuracy obtained at the expense of patient comfort. There has since been a global trend towards the development of retrospective registration methods that are non-invasive. The bulk of these developed techniques are based on intrinsic methods that only utilise the inherent information contained in a patient's image. Surface-based and intensity-based techniques are currently the most popular form of intrinsic methods, where the latter is slowly setting the standard for registration accuracy.\ud
\ud
From the literature review, it was found that surface-based registration methods are currently used the most in clinical applications. This is due to the slight speed advantage that they have over intensity-based methods. However, one of the drawbacks of surface based methods is that they cannot handle cases when the surfaces being matched significantly differ from each other. To overcome such problems requires the use of non-rigid registration techniques. However, more research is required into these approaches as the complexity involved is still too high to effectively utilise them in real-time applications. This research aims to further develop non-invasive retrospective registration techniques that are more accurate, robust and fully automatic.\ud
\ud
This report presents a thorough introduction into the field of medical image registration. It includes background on the various imaging modalities, a look at some relevant applications of registration, a classification of registration algorithms and a literature review on specific techniques. The report is then finished with a conclusion and a discussion on some future directions of registration.\u
Super-resolution for biometrics: A comprehensive survey
No full textThe lack of resolution of imaging systems has critically adverse impacts on the recognition and performance of biometric systems, especially in the case of long range biometrics and surveillance such as face recognition at a distance, iris recognition and gait recognition. Super-resolution, as one of the core innovations in computer vision, has been an attractive but challenging solution to address this problem in both general imaging systems and biometric systems. However, a fundamental difference exists between conventional super-resolution motivations and those required for biometrics. The former aims to enhance the visual clarity of the scene while the latter, more significantly, aims to improve the recognition accuracy of classifiers by exploiting specific characteristics of the observed biometric traits. This paper comprehensively surveys the state-of-the-art super-resolution approaches proposed for four major biometric modalities: face (2D+3D), iris, fingerprint and gait. We approach the super-resolution problem in biometrics from several different perspectives, including from the spatial and frequency domains, single and multiple input images, learning-based and reconstruction-based approaches. Especially, we highlight two special categories: feature-domain super-resolution which performs super-resolution directly on the feature space to purposely improve the recognition performance, and deep-learning super-resolution which discusses the most recent advances in deep learning for the super-resolution task. Finally, we discuss the current and open research challenges and provide recommendations into the future for the improved use of super-resolution with biometrics.</p
Rigid and non-rigid image registration and its association with mutual information: a review
No full textImage registration is a fundamental problem that can be found in a diverse range of fields within the\ud
research community. It is used in areas such as engineering, science, medicine, robotics, computer vision\ud
and image processing, which often require the process of developing a spatial mapping between sets of\ud
data. Registration plays a crucial role in the medical imaging field where continual advances in imaging\ud
modalities, including MRI, CTI and PET, allow the generation of 3D images that explicitly outline\ud
detailed in vivo information of not only human anatomy, but also human function.\ud
A common task within the medical imaging \ud
field is the fusing of the complimentary and synergistic\ud
information provided by the various imaging modalities. This process is known as multimodal registration. Another common task is the registration of images of the same patient taken at different times\ud
and/or in different positions. This process is referred to as mono-modal registration and can be used to\ud
track any pathological evolution. Other applications include inter-patient registration and patient-atlas\ud
matching. The \ud
first two applications are generally solved with rigid registrations, i.e. only rotations and\ud
translations are used in the transformation. However the last two examples are generally performed with\ud
a non-rigid registration. This allows one image to be deformed to match another in order to account for\ud
the non-linear local anatomic variations that exist between the images.\ud
Mutual information (MI) is a popular entropy-based similarity measure which has recently experienced\ud
a prolific expansion in a number of image registration applications. Stemming from information theory,\ud
this measure generally outperforms most other intensity-based measures in multimodal applications as it\ud
only assumes a statistical dependence between images. Introduced in the computer vision field in 1995\ud
the basic concept behind its approach is to \ud
nd a transformation, which when applied to an image, will\ud
maximise the MI between two images.\ud
The power and versatility of this measure has been demonstrated many times in the literature and\ud
consequently, is now being routinely used in clinical applications. However, despite the success and\ud
popularity of its use, it has been shown that there are cases when maximising the MI measure will lead\ud
to incorrect spatial alignments. This may be due to the presence of local or spurious global extrema\ud
which may be a result of several factors including interpolation artifacts, small image overlap, or the\ud
absence of adequate spatial correlation in the images. As a result, ongoing research into improving the\ud
robustness of this measure is still continuing. This includes the investigation of hierarchical approaches,\ud
normalisation of MI, multi-variate MI, incorporation of spatial information, along with many other\ud
optimisation, algorithmic, and implementation issues.\ud
MI has also recently found use in the non-rigid domain as often there exists a need to compute a non-rigid\ud
multimodal registration. A prominent example is in the registration of pre-operative and intra-operative\ud
images. This allows the display of pre-operative anatomical and pathological tissue discrimination in\ud
the interventional \ud
field. There have been numerous methods proposed for incorporating the MI measure\ud
into a non-rigid registration. The most obvious distinction is whether the MI is calculated in a global or\ud
local manner. There are also many ways of computing the smoothness of the deformation field. Most\ud
methods however, ensure smoothness of the deformation \ud
field by altering of the vector \ud
eld and/or by\ud
regularisation terms to constrain local deformations.\ud
This report presents a thorough introduction into the \ud
eld of medical image registration and its association with MI. This includes a general overview of all registration techniques, a more in depth look\ud
at the original MI measure and its extensions proposed in the rigid domain, an overview of non-rigid\ud
registration techniques, and \ud
nally a look at the use of MI in the non-rigid domain. On the whole, MI\ud
has proved to be a very successful measure and will no doubt be a significant aspect in image registration\ud
for years to come
Hybrid non-rigid image registration using mutual information and the viscous fluid algorithm\ud
No full textRecently there has emerged a need to compute multimodal\ud
non-rigid registrations in a lot of clinical applications. To\ud
date, the viscous fluid algorithm is perhaps the most adept\ud
method at recovering large local mis-registrations that exist\ud
between two images. However, this model can only be\ud
used on images from the same modality as it assumes similar\ud
intensity values between images. This paper presents\ud
a solution to this problem by proposing a hybrid non-rigid\ud
registration using the viscous fluid algorithmand mutual information.\ud
The mutual information is incorporated via the\ud
use of a block matching procedure to generate a sparse deformation\ud
field which drives the viscous fluid algorithm.\ud
Although successful, the hybrid approach suffers from interpolation\ud
artifacts which prevent sub-voxel deformations\ud
and limit the accuracy of the algorithm
Improved stereo image matching using mutual information and hierarchical prior probabilities
No full textMutual information (MI) has shown promise as an effective\ud
stereo matching measure for images affected by radiometric\ud
distortion. This is due to the robustness of MI\ud
against changes in illumination. However, MI-based approaches\ud
are particularly prone to the generation of false\ud
matches due to the small statistical power of the matching\ud
windows. Consequently, most previous MI approaches\ud
utilise large matching windows which smooth the estimated\ud
disparity field. This paper proposes extensions to MI-based\ud
stereo matching in order to increase the robustness of the\ud
algorithm. Firstly, prior probabilities are incorporated into\ud
the MI measure in order to considerably increase the statistical\ud
power of the matching windows. These prior probabilities,\ud
which are calculated from the global joint histogram\ud
between the stereo pair, are tuned to a two level hierarchical\ud
approach. A 2D match surface, in which the match score is\ud
computed for every possible combination of template and\ud
matching window, is also utilised. This enforces left-right\ud
consistency and uniqueness constraints. These additions to\ud
MI-based stereo matching significantly enhance the algorithm’s\ud
ability to detect correct matches while decreasing\ud
computation time and improving the accuracy
Fast exact nearest neighbour matching in high dimensions using d-D Sort
Full text linkData structures such as k-D trees and hierarchical k-means trees perform very well in approximate k nearest neighbour matching, but are only marginally more effective than linear search when performing exact matching in high-dimensional image descriptor data. This paper presents several improvements to linear search that allows it to outperform existing methods and recommends two approaches to exact matching. The first method reduces the number of operations by evaluating the distance measure in order of significance of the query dimensions and terminating when the partial distance exceeds the search threshold. This method does not require preprocessing and significantly outperforms existing methods. The second method improves query speed further by presorting the data using a data structure called d-D sort. The order information is used as a priority queue to reduce the time taken to find the exact match and to restrict the range of data searched. Construction of the d-D sort structure is very simple to implement, does not require any parameter tuning, and requires significantly less time than the best-performing tree structure, and data can be added to the structure relatively efficiently
- …
