La problématique générale de cette thèse est de reconstituer la scène de fond à partir d’une séquence d’images en présence de masques d’avant-plan. Nous nous sommes intéressés aux méthodes pour détecter ce qui constitue le fond ainsi que les solutions pour corriger les parties cachées et les distor­sions géométrique et chromatique introduites lors de la photographie.Une série de processus est proposée, dont la mise en œuvre comporte dans l'ordre l’aligne­ment géométrique, le réglage chromatique, la fusion des images et la correction des défauts.Nous nous plaçons dans l’hypothèse où le fond est porté sur une surface plane. L'aligne­ment géométrique est alors réalisé par calcul de l'homographie entre une image quelconque et l’image qui sert de référence, suivi d’une interpolation bilinéaire.Le réglage chromatique vise à retrouver un même contraste dans les différentes images. Nous proposons de modéliser la mise en cor­respondance chromatique entre images par une approximation linéaire dont les para­mètres sont déterminés par les résultats de la mise en correspondance des points de contrôle (SIFT).Ces deux étapes sont suivies par une étape de fusion. Plusieurs techniques sont comparées.La première proposition est d’étendre la définition de la médiane dans l’espace vec­toriel. Elle est robuste lorsqu’il y a plus de la moitié des images qui voient les pixels d’arrière-plan. En outre, nous concevons un algorithme original basé sur la notion de clique. Il permet de détecter le plus grand nuage de pixels dans l'espace RGB. Cette approche est fiable même lorsque les pixels d’arrière-plan sont minoritaires.Lors de la mise en œuvre de ce protocole, on constate que certains résultats de fusion présentent des défauts de type flou dus à l’existence d’erreurs d’alignement géomé­trique. Nous proposons donc un traitement complémentaire. Il est basé sur une compa­raison entre le résultat de fusion et les images alignées après passage d'un filtre gaussien. Sa sortie est un assemblage des morceaux très détaillés d'image alignés qui ressemblent le plus au résultat de fusion associés.La performance de nos méthodes est éva­luée par un ensemble de données contenant de nombreuses images de qualités diffé­rentes. Les expériences confirment la fiabi­lisé et la robustesse de notre conception dans diverses conditions de photographie. / The general topic of this thesis is to reconstruct the background scene from a burst of images in presence of masks. We focus on the background detection methods as well as on solutions to geometric and chromatic distortions introduced during ph-otography. A series of process is proposed, which con­sists of geometric alignment, chromatic adjustment, image fusion and defect correction.We consider the case where the background scene is a flat surface. The geometric align­ment between a reference image and any other images in the sequence, depends on the computation of a homography followed by a bilinear interpolation.The chromatic adjustment aims to attach a similar contrast to the scene in different im­ages. We propose to model the chromatic mapping between images with linear approximations whose parameters are decided by matched pixels of SIFT .These two steps are followed by a discus­sion on image fusion. Several methods have been compared.The first proposition is a generation of typical median filter to the vector range. It is robust when more than half of the images convey the background information. Besides, we design an original algorithm based on the notion of clique. It serves to distinguish the biggest cloud of pixels in RGB space. This approach is highly reliable even when the background pixels are the minority.During the implementation, we notice that some fusion results bear blur-like defects due to the existence of geometric alignment errors. We provide therefore a combination method as a complementary step to ameli-orate the fusion results. It is based on a com-parison between the fusion image and other aligned images after applying a Gaussian filter. The output is a mosaic of patches with clear details issued from the aligned images which are the most similar to their related fusion patches.The performance of our methods is evaluated by a data set containing extensive images of different qualities. Experiments confirm the reliability and robustness of our design under a variety of photography conditions.

京都大学 / 0048 / 新制・課程博士 / 博士(人間健康科学) / 甲第19232号 / 人健博第29号 / 新制||人健||3(附属図書館) / 32231 / 京都大学大学院医学研究科人間健康科学系専攻 / (主査)教授 杉本 直三, 教授 精山 明敏, 教授 戸井 雅和 / 学位規則第4条第1項該当 / Doctor of Human Health Sciences / Kyoto University / DFAM

The ASCAT scatterometer measures the Earth surface microwave radar backscatter in order to estimate the near-surface winds over the oceans. While the spatial resolution of the conventional applications is sufficient for many purposes, other geoscience applications benefit from an improved spatial resolution. Specialized algorithms may be applied to the scatterometer data in order to reconstruct the radar backscatter on a high-resolution grid. Image reconstruction requires the spatial response function (SRF) of each measurement, which is not reported with the measurement data. To address this need, I precisely model the SRF incorporating (1) the antenna beam response, (2) the processing performed onboard ASCAT before telemetering to the ground, and (3) the Doppler shift induced by a satellite orbiting the rotating Earth. I also develop a simple parameterized model of the SRF to reduce computational complexity. The accuracy of both models is validated.Image reconstruction of the ASCAT data is performed using the modeled SRF. I discuss the spatial resolution of the reconstructed ASCAT images and consider the first- and second-order statistics of the reconstructed data. Optimum values for the parameters of the reconstruction algorithms are also considered. The reconstructed radar backscatter data may be used for enhanced-resolution wind retrieval and for geoscience applications. In this dissertation, the reconstructed backscatter data is used to map the surface extent of the 2010 Deepwater Horizon oil spill and in a study to quantify the azimuth angle anisotropy of backscatter in East Antarctica. Near-coastal ocean wind retrieval is also explored in this dissertation. Because near-coastal ocean measurements of backscatter may be “contaminated” from nearby land and introduce errors to wind retrieval, they must be discarded. The modeled SRF is used to quantify the land contamination, enabling enhanced-resolution wind retrieval much closer to the coasts. The near-coastal winds are validated against buoy measurements.

No description available.

Due to the mission-critical nature of the autonomous driving application, underlying algorithms for scene understanding should be given special care during their development. Mostly, they should be designed with precise consideration of accuracy and run-time. Accuracy should be considered strictly which if compromised leads to faulty interpretation of the environment that may ultimately result in accidental scenarios. On the other hand, run-time holds an important position as the delayed understanding of the scene would hamper the real-time response of the vehicle which again leads to unforeseen accidental cases. These factors come as the functions of several factors such as the design and complexity of the algorithms, nature of the encountered objects or events in the environment, weather-induced effects, etc. In this work, several novel scene understanding algorithms in terms- of semantic segmentation are devised. First, a transfer learning technique is proposed in order to transfer the knowledge from the data-rich domain to a data-scarce off-road driving domain for semantic segmentation such that the learned information is efficiently transferred from one domain to another while reducing run-time and increasing the accuracy. Second, the performance of several segmentation algorithms is assessed under the easy-to-severe rainy condition and two methods for achieving the robustness are proposed. Third, a new method of eradicating the rain from the input images is proposed. Since autonomous vehicles are rich in sensors and each of them has the capability of representing different types of information, it is worth fusing the information from all the possible sensors. Forth, a fusion mechanism with a novel algorithm that facilitates the use of local and non-local attention in a cross-modal scenario with RGB camera images and lidar-based images for road detection using semantic segmentation is executed and validated for different driving scenarios. Fifth, a conceptually new method of off-road driving trail representation, called Traversability, is introduced. To establish the correlation between a vehicle’s capability and the level of difficulty of the driving trail, a new dataset called CaT (CAVS Traversability) is introduced. This dataset is very helpful for future research in several off-road driving applications including military purposes, robotic navigation, etc.

AMSR-E is a space-borne radiometer which measures Earth microwave emissions or brightness temperatures (Tb) over a wide swath. AMSR-E data and images are useful in mapping valuable Earth-surface and atmospheric phenomena. A modified version of the Scatterometer Image Reconstruction (SIR) algorithm creates Tb images from the collected data. SIR is an iterative algorithm with tuning parameters to optimize the reconstruction for the instrument and channel. It requires an approximate aperture function for each channel to be effective. This thesis presents a simulator-based optimization of SIR iteration and aperture function threshold parameters for each AMSR-E channel. A comparison of actual Tb images generated using the optimal and sub-optimal values is included. Tuned parameters produce images with sharper transitions between regions of low and high Tb for lower-frequency channels. For higher-frequency channels, the severity of artifacts due to temporal Tb variation of the input measurements decreases and coverage gaps are eliminated after tuning. A two-parameter Gaussian-like bell model is currently assumed in image reconstruction to approximate the AMSR-E aperture function. This paper presents a method of estimating the effective AMSR-E aperture function using Tb measurements and geographical information. The estimate is used as an input for image reconstruction. The resulting Tb images are compared with those produced with the previous Gaussian approximation. Results support the estimates found in this paper for channels 1h, 1v, and 2h. Images processed using the old or new aperture functions for all channels differed by a fraction of a Kelvin over spatially smooth regions.

In the infection process, native poliovirus (160S) first converts to a cell-entry intermediate (135S) particle, which causes the externalization of capsid proteins VP4 and the N-terminus of VP1 (residues 1-53). The externalization of these entities is followed by release of the RNA genome, leaving an empty (80S) particle. Three antibodies were utilized to track the location of VP1 residues in different states of poliovirus by cryogenic electron microscopy (cryo-EM). "P1" antibody binds to N-terminal residues 24-40 of VP1. Three-dimensional reconstruction of 135S-P1 showed that P1 binds to a prominent capsid peak known as the "propeller tip". In contrast, our initial 80S-P1 reconstruction showed P1 Fabs also binding to a second site, ~60 Å distant, at the icosahedral twofold axes. Analysis of 80S-P1 reconstructions showed that the overall population of 80S-P1 particles consisted of three kinds of capsids: those with P1 Fabs bound only at the propeller tips; only at the twofold axes; or simultaneously at both positions. Our results indicate that, in 80S particles, a significant fraction of VP1 can deviate from icosahedral symmetry. Similar deviations from icosahedral symmetry may be biologically significant during other viral transitions. "C3" antibody binds to 93-103 residues (BC loop) of VP1. The C3 epitope shifts outwards in radius by 4.5% and twists through 15° in the 160S-to-135S transition, but appears unchanged in the 135S-to-80S transition. In addition, binding of C3 to either 160S or 135S particles causes residues of the BC loop to move an estimated 5 (±2) Å, indicating flexibility. The flexibility of BC loop may play a role in cell-entry interactions. At 37°C, the structure of poliovirus is dynamic, and internal polypeptides VP4 and the N-terminus of VP1 externalize reversibly. An antibody, binding to the residues 39-55 of VP1, was utilized to track the location of the N-terminus of VP1 in 160S particle in the "breathing" state. The resulting reconstruction showed the capsid expands similarly to the irreversibly altered 135S particle, but the N-terminus of VP1 is located near the twofold axes, instead of the propeller tip as in 135S particles.

Factorized backprojection is a processing algorithm for reconstructing images from data collected by synthetic aperture radar (SAR) systems. Factorized backprojection requires less computation than conventional time-domain backprojection with little loss in accuracy for straight-line motion. However, its implementation is not as straightforward as direct backprojection. Further, implementing an azimuth window has been difficult in previous versions of factorized backprojection. This thesis provides a new, easily parallelizable formulation of factorized backprojection designed for both pulsed and linearly frequency modulated continuous wave (LFM-CW) stripmap SAR data. A method of easily implementing an azimuth window as part of the factorized backprojection algorithm is introduced. The approximations made in factorized backprojection are investigated and a detailed analysis of the corresponding errors is provided. We compare the performance of windowed factorized backprojection to direct backprojection for simulated and actual SAR data.

We propose an exact filtered backprojection algorithm for inversion of the cone beam data in the case when the trajectory is composed of a distorted circle and a line segment. The length of the scan is determined by the region of interest , and it is independent of the size of the object. With few geometric restrictions on the curve, we show that we have an exact reconstruction. Numerical experiments demonstrate good image quality.

Monte Carlo is an important and well established research tool used in emission tomography. While used extensively in research applications, these techniques are not typically implemented clinically due to their low detection efficiency and long acquisition times. In order to make this computational tool faster, the variance reduction technique known as convolution-based forced detection (CFD) has been implemented into the SIMIND MC code (CFD-SIMIND) by our group. Briefly, at each site of interaction within the object, photons are forced to travel in a direction perpendicular to the detector and are then convolved with a distance dependent blurring kernel specific to that collimator and photon energy. A similar CFD method has already been implemented as an option in the SIMIND Monte Carlo program. The study presented in Chapter 2 performs a comparison between a well established, non-VRT Monte Carlo program, GATE, with our accelerated CFD-SIMIND. The intent of this work is to establish if CFD-SIMIND can either replace or be used in conjunction with GATE in order to gain significant reduction in simulation times for low and medium energy isotopes. A number of simulation studies were performed using point sources in air and water, along with the 3D XCAT phantom and a rectangular sheet source for Tc-99m with low and medium energy collimator and In-111 with medium energy collimator. A comparison in the projection domain was then performed in terms of spatial resolution, sensitivity, image profiles and energy spectra. The study has shown percent differences of between 3−5% in sensitivity between CFD-SIMIND and GATE with mean universal image quality index value of 0.994 ± 0.009 and spatial resolution within 0.2 mm of each other. CFD-SIMIND offers a significant reduction in simulation time by a factor of 5−6 orders of magnitude compared to GATE. This acceleration time is useful for many applications. This study also provides an objective tool that can help to determine if CFD-SIMIND can be used in place of GATE in order to achieve images of sufficient quality within a reduced time and at much lower computational cost. Simultaneous multi-isotope SPECT imaging has a number of applications in cardiac, brain and cancer imaging. The major concern however, is the significant crosstalk contamination due to photon scatter between the different isotopes. The second study (Chapter 3) focuses on a method of downscatter compensation between two isotopes iii in simultaneous dual isotope SPECT acquisition applied to cancer imaging using Tc-99m and In-111. We have developed an iterative image reconstruction technique that simulates the photon down-scatter from one isotope into the acquisition window of a second isotope. Our approach uses CFD-SIMIND for the forward projection step in an iterative reconstruction algorithm. The MC estimated scatter contamination of a radionuclide contained in a given projection view is then used to compensate for the photon contamination in the acquisition window of other nuclide. We use a modified ordered subset-expectation maximization (OS-EM) algorithm named simultaneous ordered subset-expectation maximization (Sim-OSEM), to perform this step. In this study, we have undertaken a number of simulation tests and phantom studies to verify this approach. The proposed reconstruction technique was also evaluated by reconstruction of experimentally acquired phantom data. Reconstruction using Sim- OSEM showed very promising results in terms of contrast recovery and uniformity of object background compared to alternative reconstruction methods implementing alternative scatter correction schemes (i.e., triple energy window or separately acquired projection data). In this study the evaluation is based on the quality of reconstructed images and activity estimated using Sim-OSEM. In order to quantitate the possible improvement in spatial resolution and signal to noise ratio (SNR) observed in this study, further simulation and experimental studies are required. It is perceived that in simultaneous dual-isotope breast SPECT studies using 123I-labelled Z-MIVE and Tc-99m sestamibi, I-123 labelled Z-MIVE not only detects the presence of estrogen receptor (ER) but, also thought to complement Tc-99m sestamibi in differentiating between benign and malignant breast lesions for patients with breast cancer (Chapter 4). The major concern in simultaneous Tc-99m/I-123 SPECT is the significant crosstalk contamination between the different isotopes used. The current study focuses on a method of crosstalk (downscatter and spillover) compensation between two isotopes with data acquired using Thallium activated Sodium Iodide (NaI(Tl)) detector (Energy resolution 9.8% at 140 keV ) and Cadmium Zinc Telluride (CZT) detector (Energy resolution 5% 140 keV ) respectively. The study uses Sim- OSEM for crosstalk compensation between the isotopes. We have undertaken a number of simulation studies using our modeled breast phantom to verify this approach. Reconstruction using Sim-OSEM showed very promising results in terms of crosstalk and scatter compensation and uniformity of background. In our case images obtained using Sim-OSEM were comparable or even better than the images reconstructed from separately acquired projection data using analytical attenuation based reconstruction. This may be due to better small angle scatter compensation in case of Sim-OSEM as CFD-SIMIND based MC forward projector was used. Compensation of the image degradation effects (i.e. attenuation, scatter and collimator-detector response) is necessary for an accurate quantification in SPECT imaging. We have previously proposed an accelerated Monte Carlo (MC) based iterative SPECT reconstruction algorithm that accurately corrects for attenuation and scatter once provided with attenuation information (Chapters 3 and 4). This algorithm uses SIMIND MC program accelerated through the implementation of a variance reduction technique known as, convolution forced detection (CFD), (CFD-SIMIND). With ever increasing number of hybrid SPECT/CT systems, CT-based attenuation correction is becoming a standard clinical protocol. This co-registered CT image with SPECT data can also be used to incorporate anatomical information as a prior into a maximum a-posteriori (MAP) SPECT image reconstruction algorithm. The study presented in Chapter 5 proposes a MAP reconstruction algorithm that includes CFD-SIMIND as a forward projector and a CT-image as an anatomical prior (CFD-AMAP) for simultaneous compensation of scatter and attenuation and, enhancement of spatial resolution during reconstruction. We have performed a number of simulation and experimental studies to elaborate the advantages of CFD-AMAP. These studies show an accurate quantification (within ±5% and ±8% for simulation and experimental studies respectively) accompanied by a significant reduction in coefficient of variation (CoV ). This reduction of CoV results in an improved boundary delineation and the Gibbs artifact compensation. However, this compensation comes at the cost of loss of an overall contrast in the reconstructed images due to a more uniform distribution of estimated activity over the regions of interest (ROI’s). Further studies with more complex phantoms and real patient data, task-based ROC studies, improvement in CFD-SIMIND in terms of speed and use of better Bayesian image reconstruction algorithms are needed to elaborate on the strengths and weaknesses of this proposed MC based forward projector and to pave the way for CFD-SIMIND based image reconstruction algorithms from research to clinic. / Thesis / Doctor of Philosophy (PhD)