Digital inpainting is the reconstruction of a missing or damaged region in a digital image. Intensity values in the missing region are approximated using information near the boundary of the region. Some applications include repair of chipped paintings, repair of rips in paper photographs, and removal of unwanted objects from photographs. In this thesis, we review 2D digital inpainting techniques, examine the application of 3D digital inpainting to cell path reconstruction, and propose a new inpainting technique inspired by the cell path reconstruction problem. Cell path reconstruction is the estimation of the shape and position of living cells in videos recorded using fluorescence microscopy. This procedure is necessary because in a particular phase of the life cycle of some cells, fluorescent light passes through the cells with an undetectable change in wavelength and they vanish from the frame. This leads to misleading results when, for example, the number of cells in a particular frame is counted. We transform the position/shape estimation problem into a 3D shape reconstruction problem by stacking the frames of the video to form a 3D volume. In this volume, cell paths form tubes with missing segments where cells have vanished. We apply elastica inpainting to the 3D tube reconstruction problem and introduce a new 3D inpainting model to overcome difficulties with a direct generalization to 3D of 2D elastica.

Digital inpainting is the reconstruction of a missing or damaged region in a digital image. Intensity values in the missing region are approximated using information near the boundary of the region. Some applications include repair of chipped paintings, repair of rips in paper photographs, and removal of unwanted objects from photographs. In this thesis, we review 2D digital inpainting techniques, examine the application of 3D digital inpainting to cell path reconstruction, and propose a new inpainting technique inspired by the cell path reconstruction problem. Cell path reconstruction is the estimation of the shape and position of living cells in videos recorded using fluorescence microscopy. This procedure is necessary because in a particular phase of the life cycle of some cells, fluorescent light passes through the cells with an undetectable change in wavelength and they vanish from the frame. This leads to misleading results when, for example, the number of cells in a particular frame is counted. We transform the position/shape estimation problem into a 3D shape reconstruction problem by stacking the frames of the video to form a 3D volume. In this volume, cell paths form tubes with missing segments where cells have vanished. We apply elastica inpainting to the 3D tube reconstruction problem and introduce a new 3D inpainting model to overcome difficulties with a direct generalization to 3D of 2D elastica.

Segmentation of an image into individual tree crowns is a key step in the processing of remotely sensed data for forestry practice. The doctoral thesis gives a broad overview of this topic. It comprehends theoretical context from mathematical point of view and defines basic terms from airborne imaging and laser scanning. Mathematical methods of tree detection are focused on a robust adaptation to the actual conditions in a region of interest. A novel approach of crown area delineation is introduced, it combines a seeded region growing technique with an active contour as a crown boundary representation. The parametrisation of all algorithms is analysed in a practical half of the thesis and more application-oriented issues are mentioned. Executable computer programs are attached.

La mesure simultanée des champs cinématiques et thermiques est très importante pour les procédures thermomécaniques. Les caméras à base de silicium sont largement utilisées pour l'observation en temps réel des champs cinématiques, principalement grâce à la corrélation d'images numériques. De plus, ils sont aussi connus pour sa sensibilité dans le spectre du proche infrarouge, ce qui permet d’acquérir des champs thermiques à l’aide d’une caméra à base de silicium. Cependant, pour la caméra à base de silicium, il y a deux problèmes principaux d’obtenir simultanément des champs cinématiques et thermiques. D’abord, dans le spectre du proche infrarouge, une petite variation de température entraînera une modification importante du niveau de gris de l'image, ce qui entraînera facilement une mauvaise qualité des images. Deuxième, la corrélation d’images numériques nécessite une surface hétérogène et contrastée, tandis que la thermographie dans le proche infrarouge nécessite une surface homogène et constante. Dans cette thèse, une technique innovante a été proposée pour ajuster automatiquement le temps d'exposition de la caméra afin d'obtenir des images exploitables pour l’analyse cinématique et thermique, quel que soit l'évolution de température à la surface de l'objet observé. Cette technique a été validée par expériences différentes, notamment des expériences de chauffage d’un corps noir et des expériences de chauffage d’un échantillon réel. Les modèles radiométriques du corps noir et de la surface des échantillons calibrent respectivement. Basé sur les modèles radiométriques, des champs thermiques ont été reconstruits sur les images exploitables pour l’analyse cinématique et thermique. L'expérience à haute température est réalisée pour le ballonnement des tubes où les champs cinématiques et thermiques sont observés. La corrélation d'images numériques a été effectuée globalement afin d'obtenir des champs cinématiques. Pour effectuer la thermographie du proche infrarouge sur la surface de l’échantillon, le modèle radiométrique est étalonné selon une partie des pixels les plus brillants. Dans ce cas, 20% des pixels les plus brillants sont utilisés pour effectuer l'étalonnage des modèles radiométriques. Basée sur le modèle en utilisant 20% des pixels plus brillants, les champs thermiques sont reconstruits. Combiné avec les coordonnées connues du champ cinématique par corrélation d'images numériques, le champ thermique et le champ cinématique dans les mêmes coordonnées peut être obtenu. / Simultaneous measurement of kinematic and thermal full fields are very important for thermomechanical procedures. Silicon-based cameras are widely used to perform real-time observation of the kinematic fields, mainly thanks to digital image correlation. Moreover, they are known to be as well sensitive in the near-infrared spectral range, thus the acquirement of thermal fields using silicon-based cameras is possible. However, there are two main problems for the silicon-based camera to obtain simultaneously kinematic and thermal fields. One is that in the near-infrared spectral range, a small temperature variation will lead to a large modification in the image gray level, which easily leads to poor quality images. Another is that digital image correlation needs a heterogeneous and contrasting surface, while the near-infrared thermography needs a homogeneous and constant surface. In this thesis, an innovative technique was proposed to automatically adjust the exposure time of the camera to obtain kinematically and thermally exploitable images whatever the temperature evolution occurs on the surface of the observed object. This technique was validated by different experiments, including blackbody heating experiments and realistic specimen heating experiments. Radiometric models of blackbody and specimen surfaces ware calibrated respectively. Based on the radiometric models, thermal fields have been reconstructed on the kinematically and thermally exploitable images. High temperature tube ballooning experiment is conducted to perform both kinematic and thermal fields. Global digital image correlation was performed to obtain kinematic fields. To perform near-infrared thermography on the specimen surface, radiometric model is calibrated based on portions of the brightest pixels. In this case 20% of the brightest pixels are used to perform radiometric model calibration. Based on the radiometric model using 20% of the brightest pixels, the thermal fields are reconstructed. Combined with the known coordinates of kinematic fields by digital image correlation, the thermal fields at the same coordinates as kinematic fields can be obtained.