Ce travail porte sur la modélisation de problèmes liés au recalage d'images. Le recalage consiste à trouver une déformation optimale de sorte qu'une image déformée s'aligne sur une image de référence. Il s'agit d'une technique que l'on rencontre dans de nombreux domaines, comme l'imagerie médicale, la comparaison de données ou le suivi de formes. Le premier chapitre se concentre sur le problème de préservation de la topologie. Cette condition de préservation de la topologie est importante lorsque la déformation recherchée traduit des propriétés physiques des objets soumis à la déformation. Les chapitres suivants proposent la construction de différentes méthodes de recalage d'images fondées sur la théorie de l'élasticité non linéaire. En effet, les objets à apparier sont supposés être des matériaux hyper-élastiques. Différents termes d'attaches aux données ont été explorés ainsi que deux modèles conjoints de segmentation et recalage. / This work focuses on the modelling of problems related to image registration. Image registration consists in finding an optimal deformation such that a deformed image is aligned with a reference image. It is an important task encountered in a large range of applications such as medical imaging, comparison of data or shape tracking. The first chapter concerns the problem of topology preservation. This condition of topology preservation is important when the sought deformation reflects physical properties of the objects to be distorted. The following chapters propose several methods of image registration based on the nonlinear elasticity theory. Indeed, the objects to be matched are modelled as hyperelastic materials. Different fidelity terms have been investigated as well as two joint segmentation/registration models.

In the application of stereotactic radiosurgery for treatment of brain tumors it is imperative that patients are positioned with sub-millimetre accuracy so as to not damage surrounding healthy tissue during treatment. We investigate how feasible the technique of digitally reconstructed radiographs is as a registration method for patient positioning in this type of high-accuracy application. In particular, since most registration methods based on said technique only rely on two simultaneous projection angles seldom reaching high enough accuracy, we consider any arbitrary amount and observe its effects on the registration accuracy. Three different approaches are considered for how multiple projections can be combined into one single metric. The results seem to indicate that although computed tomography yields accuracies well within the boundaries for stereotactic radiosurgery, cone beam computed tomography in its current state does not. Possible reasons explaining the difference include problems with reconstruction artifacts in the model and inadequate metrics. / Inom tillämpningen av stereotaktisk strålkirurgi för behandling av hjärntumörer är det absolut nödvändigt att patienter positioneras inom millimeternoggrannhet för att undvika skada hos närliggande vävnad vid behandling. Vi undersöker hur trolig tekniken om digitaliserade återskapande röntgenbilder är som en registreringsmetod för patientpositionering i den här typen av mycket noggranna applikationer. Framförallt betraktas godtyckligt många projektionsvinklar och observerar dess effekt på registreringsnoggrannheten, eftersom de flesta tidigare presenterade registreringsmetoderna endast är baserade på två projektionsvinklar och sällan uppfyller det krav som ställs på noggrannhet. Tre olika tillvägagångssätt betraktas för hur flertalet projektionsvinklar kan kombineras till ett enda avståndsmått. Resultaten verkar tyda på att datortomografi medför noggrannheter väl inom ramen för stereotaktisk strålkirurgi, medan konstråledatortomografi i sitt nuvarande tillstånd inte verkar göra detta. Anledningar till denna olikhet tros vara problem med rekonstruktionsartefakter i modellen och otillräckliga avståndsmått.

Over 30 000 liver, abdomen, and thigh slices have been acquired by computed tomography for the SCAPIS and IGT study. To utilise the full potential of the large cohort and enable statistical pixel-wise body composition analysis and visualisation of associations with other biomarkers, a point-to-point correspondence between the scans is needed. For this purpose, an inter-subject image registration pipeline that combines the low-level information from CT images with high-level information from segmentation masks have been developed. It uses tissue-specific regularisation and processes images efficiently. The pipeline was used to deform 4603 CT scans of each slice into a respective common reference space in less than 30 hours. All but the ribs in the liver slices and the intra abdominal region of the abdomen were generally registered correctly. Vector and intensity magnitude errors indicating inverse consistency were on average less than 2.5 mm and 40 Hounsfield units respectively. The method may serve as a starting point for statistical pixel-wise body composition analysis, its association with non-imaging data, as well as saliency mapping analysis of the three-slice CT scans from the large SCAPIS and IGT cohorts.

This thesis tackles the challenge of registering modern to historic images in the context of urban rephotography. It aims at automatically identifying stable image features in scenes, which have been exposed to medium to tremendous changes across the years. Instead, the related field of location recognition mainly focuses on illumination and seasonal changes. This work illustrates that common feature descriptors are applicable in the context of historic and modern image matching, while local detectors are not, but most important is the choice of appropriate correspondence filters. It is verified that major structural changes are most challenging for traditional image matching approaches and the methods developed in this work are applicable to challenging image pairs beyond rephotography. Besides, features extracted from Convolutional Neural Networks (CNNs), originally trained for the task of location recognition, show high performance and should be further developed for the specific task of historic to modern image matching. At last, practical developments are presented, including an online portal for presenting and organizing rephotographs as well as an initial version of a mobile application, which supports recovering the original viewpoint of an image.

With advancements in space technology, remote sensing applications, and computer vision, significant improvements in the data describing our planet are seen today. Researchers want to gather different kinds of data and perform data fusion techniques between them to increase our understanding of the world. Two such data types are Electro-Optical images and Synthetic Aperture Radar images. For data fusion, the images need to be accurately aligned. Researchers have investigated methods for robustly and accurately registering these images for many years. However, recent advancements in imaging systems have made the problem more complex than ever. Currently, the imaging satellites that capture information around the globe have achieved a resolution of less than a meter per pixel. There is an increase in signal complexity for high-resolution SAR images due to how the imaging system operates. Interference between waves gives rise to speckled noise and geometric distortions, making the images very difficult to interpret. This directly affects the image registration accuracy. In this thesis, the complexity of the problem regarding registration between SAR and EO data was described, and methods for registering the images were investigated. The methods were feature- and area-based. The feature-based method used a KAZE filter and SURF descriptor. The method found many key points but few correct correspondences. The area-based methods used FFT and MI, respectively. FFT was deemed best for higher quality images, whereas MI better dealt with the non-linear intensity difference. More complex techniques, such as dense neural networks, were excluded. No method achieved satisfying results on the entire data set, but the area-based methods accomplished complementary results. A conclusion was drawn that the distortions in the SAR images are too significant to register accurately using only CV algorithms. Since the area-based methods achieved good results on images excluding significant distortions, future work should focus on solving the geometrical errors and increasing the registration accuracy

This thesis deals with image registration of MRI images using neural networks. Image registration combines multiple images of the same subject that were taken at different points in time, from different sensors, or from different points of views into a single image and coordinate system. Image registration is widely used in medical imaging and remote sensing. In this thesis feed forward neural networks and wavelet neural networks are used to estimate the parameters of registration. Simulations show that the wavelet networks provide significantly more accurate results than feed forward networks and other proposed methods including genetic algorithms. Both methods are also shown to be robust to noise and changes in parameter ranges.

This thesis is concerned with model-based non-rigid registration of single-modality magnetic resonance images of compressed and uncompressed breast tissue in breast cancer diagnostic/interventional imaging. First, a volumetric registration algorithm is developed which solves the registration as a state estimation problem. Using a static deformation model. To reduce computations, the similarity measure is calculated at some specific points called control points. These control points can be from a low resolution image grid or any irregular image grid. Our numerical analysis has shown that control points placed in the area without much information; i.e with small or no changes in image intensity, yield negligible deformation. Therefore, the selection of the control points can significantly impact the accuracy and computation complexity of the registration algorithms. An extension of the speeded up robust features (SURF) to 3D is proposed for enhanced selection of the control points in deformable image registration. The impact of this new control point selection method on the performance of the registration algorithm is analyzed by comparing it to the case where regular grid control points are used. The results show that the number of control points could be reduced by a factor of ten with new selection methodology without sacrificing performance. Second image registration method is proposed in which, based on a segmented pre-operative image, a deformation model of the breast tissue is developed and discretized in the spatial domain using the method of finite elements. The compression of the preoperative image is modeled by applying smooth forces on the surface of the breast where compression plates are placed. Image registration is accomplished by formulating and solving an optimization problem. The cost function is a similarity measure between the deformed preoperative image and intra-operative image computed at some control point and the decision variables are the tissue interaction forces. / Thesis / Master of Applied Science (MASc)

Image-guided radiation therapy (IGRT) provides accurate and precise tumour targeting. To ensure adequate coverage in IGRT, a planning target volume (PTV) margin is added around the target to account for treatment uncertainties. Treatment plans are designed to deliver a high percentage of the prescription dose to the PTV; thus, portions of healthy tissue are also subjected to high radiation dose. IGRT employs dedicated devices that enable visual assessment of some treatment uncertainties, such as variations in patient set-up. Safe and effective IGRT delivery requires adherence to disease site-specific protocols that describe process details such as imaging technique, alignment method, and corrective action levels. Protocol design is challenging since its effect on treatment accuracy is currently unknown. This thesis aims to understand the interplay between lung IGRT protocol parameters by developing a framework that quantifies geometrical accuracy. Deformable image registration was used to account for changes in target shape and size throughout treatment. Sufficient accuracy was considered when at least 99% of the target surface fell within the PTV. This analysis revealed that the clinical 10 mm PTV margin can be safely reduced by at least 2 mm in each direction. Evaluation of IGRT accuracy was extended to spinal cord alignment. Simulations were carried out with various matching strategies to correct for set-up error, including rotational off-sets. Inappropriate combinations of matching strategies and safety margins resulted in sub-optimal geometrical coverage. Various lung IGRT protocol options were recommended to optimize accuracy and workflow efficiency. For example, an 8 mm PTV margin can be used with spinal cord alignment, a 4 mm cord margin, and up to 5° of rotational error. A more aggressive protocol involved a 6 mm PTV margin with direct target alignment, a 5 mm cord margin, and a 4° rotational tolerance. / Thesis / Master of Science (MSc)

Object Tracking plays a very pivotal role in many computer vision applications such as video surveillance, human gesture recognition and object based video compressions such as MPEG-4. Automatic detection of any moving object and tracking its motion is always an important topic of computer vision and robotic fields. This thesis deals with the problem of detecting the presence of debris or any other unexpected objects in footage obtained during spacecraft launches, and this poses a challenge because of the non-stationary background. When the background is stationary, moving objects can be detected by frame differencing. Therefore there is a need for background stabilization before tracking any moving object in the scene. Here two problems are considered and in both footage from Space shuttle launch is considered with the objective to track any debris falling from the Shuttle. The proposed method registers two consecutive frames using FFT based image registration where the amount of transformation parameters (translation, rotation) is calculated automatically. This information is the next passed to a Kalman filtering stage which produces a mask image that is used to find high intensity areas which are of potential interest.

No description available.