No description available.

<p>Imaging-based analysis of developmental processes are crucial to understand the mechanisms controlling plant and animal development. In vertebrate embryos such as the zebrafish embryo, nuclei segmentation plays an important role to detect and quantify nuclei over space and time. However, limitations of the image quality and segmentation methods may affect the segmentation performance. In plant including studies on Arabidopsis epidermis growth, cellular shape change dictates organ size control and growth behavior, and quantitative image analysis of dynamics cell patterning is needed to link the cause and effect between cells and organs. Here we provide a series of new quantitative biological imaging methods a series of new quantitative biological imaging methods and tools including wavelet-based segmentation method in zebrafish embryo development studies and convex hull approach for quantitative shape analyses of lobed plant cells.</p> <p> </p> <p>Identification of individual cells in tissues, organs, and in various developing systems is a well-studied problem because it is an essential part of objectively analyzing quantitative images in numerous biological contexts. In this paper we present a size dependent wavelet-based segmentation method that provides robust segmentation without any preprocessing, filtering or fine-tuning steps, and is robust to the signal-to-noise ratio (SNR). The program separates overlapping nuclei, identifies cell cycle states and minimizes intensity attenuation in object identification. The wavelet-based methods presented herein achieves robust segmentation results with respect to True Positive rate, Precision, and segmentation accuracy compared with other commonly used methods. We applied the segmentation program to Zebrafish embryonic development IN TOTO quantification and developed an automatic interactive imaging analysis platform named WaveletSEG, that integrates nuclei segmentation, image registration, and nuclei shape analysis. A set of additional functions we developed include a 3D ground truth annotation tool, a synthetic image generator, a segmented training datasets export tool, and data visualization interfaces are also incorporated in WaveletSEG for additional data analysis and data validation. </p> <p> </p> <p>In addition to our work in Zebrafish, we developed image analysis tools for quantitative studies of cell-to-organ in plants. Given the importance of the epidermis and this particular cell type for leaf expansion, there is a strong need to understand how pavement cells morph from a simple polyhedral shape into highly lobed and interdigitated cells. Currently, it is still unclear how and when patterns of lobing are initiated in pavement cells, and one major technological bottleneck to address the problem is the lack of a robust and objective methodology to identify and track lobing events during the transition from simple cell geometry to lobed cells. We develop a convex-hull-based algorithm termed LobeFinder to identify lobes, quantify geometric properties, and create a useful graphical output for further analysis. The algorithm is validated against manually curated cell images of pavement cells of widely varying sizes and shapes. The ability to objectively count and detect new lobe initiation events provides an improved quantitative framework to analyze mutant phenotypes, detect symmetry-breaking events in time-lapse image data, and quantify the time-dependent correlation between cell shape change and intracellular factors that may play a role in the morphogenesis process.</p>

This thesis explores motion trajectory-based approaches to recognize human actions in real-world, unconstrained videos. Recognizing actions is an important task in applications such as video retrieval, surveillance, human-robot interactions, analysis of sports videos, summarization of videos, behaviour monitoring, etc. There has been a considerable amount of research done in this regard. Earlier work used to be on videos captured by static cameras where it was relatively easy to recognise the actions. With more videos being captured by moving cameras, recognition of actions in such videos with irregular camera motion is still a challenge in unconstrained settings with variations in scale, view, illumination, occlusion and unrelated motions in the background. With the increase in videos being captured from wearable or head-mounted cameras, recognizing actions in egocentric videos is also explored in this thesis. At first, an effective motion segmentation method to identify the camera motion in videos captured by moving cameras is explored. Next, action recognition in videos captured in normal third-person view (perspective) is discussed. Further, the action recognition approaches for first-person (egocentric) views are investigated. First-person videos are often associated with frequent unintended camera motion. This is due to the motion of the head resulting in the motion of the head-mounted cameras (wearable cameras). This is followed by recognition of actions in egocentric videos in a multicamera setting. And lastly, novel feature encoding and subvolume sampling (for “deep” approaches) techniques are explored in the context of action recognition in videos. The first part of the thesis explores two effective segmentation approaches to identify the motion due to camera. The first approach is based on curve fitting of the motion trajectories and finding the model which best fits the camera motion model. The curve fitting approach works when the trajectories generated are smooth enough. To overcome this drawback and segment trajectories under non-smooth conditions, a second approach based on trajectory scoring and grouping is proposed. By identifying the instantaneous dominant background motion and accordingly aggregating the scores (denoting the “foregroundness”) along the trajectory, the motion that is associated with the camera can be separated from the motion due to foreground objects. Additionally, the segmentation result has been used to align videos from moving cameras, resulting in videos that seem to be captured by nearly-static cameras. In the second part of the thesis, recognising actions in normal videos captured from third-person cameras is investigated. To this end, two kinds of descriptors are explored. The first descriptor is the covariance descriptor adapted for the motion trajectories. The covariance descriptor for a trajectory encodes the co-variations of different features along the trajectory’s length. Covariance, being a second-order encoding, encodes information of the trajectory that is different from that of the first-order encoding. The second descriptor is based on Granger causality. The novel causality descriptor encodes the “cause and effect” relationships between the motion trajectories of the actions. This type of interaction descriptors captures the causal inter-dependencies among the motion trajectories and encodes complimentary information different from those descriptors based on the occurrence of features. The causal dependencies are traditionally computed on time-varying signals. We extend it further to capture dependencies between spatiotemporal signals and compute generalised causality descriptors which perform better than their traditional counterparts. An egocentric or first-person video is captured from the perspective of the personof-interest (POI). The POI wears a camera and moves around doing his/her activities. This camera records the events and activities as seen by him/her. The POI who is performing actions or activities is not seen by the camera worn by him/her. Activities performed by the POI are called first-person actions and third-person actions are those done by others and observed by the POI. The third part of the thesis explores action recognition in egocentric videos. Differentiating first-person and third-person actions is important when summarising/analysing the behaviour of the POI. Thus, the goal is to recognise the action and the perspective from which it is being observed. Trajectory descriptors are adapted to recognise actions along with the motion trajectory ranking method of segmentation as pre-processing step to identify the camera motion. The motion segmentation step is necessary to remove unintended head motion (camera motion) during video capture. To recognise actions and corresponding perspectives in a multi-camera setup, a novel inter-view causality descriptor based on the causal dependencies between trajectories in different views is explored. Since this is a new problem being addressed, two first-person datasets are created with eight actions in third-person and first-person perspectives. The first dataset is a single camera dataset with action instances from first-person and third-person views. The second dataset is a multi-camera dataset with each action instance having multiple first-person and third-person views. In the final part of the thesis, a feature encoding scheme and a subvolume sampling scheme for recognising actions in videos is proposed. The proposed Hyper-Fisher Vector feature encoding is based on embedding the Bag-of-Words encoding into the Fisher Vector encoding. The resulting encoding is simple, effective and improves the classification performance over the state-of-the-art techniques. This encoding can be used in place of the traditional Fisher Vector encoding in other recognition approaches. The proposed subvolume sampling scheme, used to generate second layer features in “deep” approaches for action recognition in videos, is based on iteratively increasing the size of the valid subvolumes in the temporal direction to generate newer subvolumes. The proposed sampling requires lesser number of subvolumes to be generated to “better represent” the actions and thus, is less computationally intensive compared to the original sampling scheme. The techniques are evaluated on large-scale, challenging, publicly available datasets. The Hyper-Fisher Vector combined with the proposed sampling scheme perform better than the state-of-the-art techniques for action classification in videos.

Made available in DSpace on 2014-12-17T14:55:58Z (GMT). No. of bitstreams: 1 PauloFB.pdf: 617080 bytes, checksum: 5f9c06266e137d13c8b810163ede788d (MD5) Previous issue date: 2005-08-25 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior / Microstrip antennas are widely used in modern telecommunication systems. This is particularly due to the great variety of geometries and because they are easily built and integrated to other high frequency devices and circuits. This work presents a study of the properties of the microstrip antenna with an aperture impressed in the conducting patch. Besides, the analysis is performed for isotropic and anisotropic dielectric substrates. The Multiport Network Model MNM is used in combination with the Segmentation Method and the Greens function technique in the analysis of the considered microstrip antenna geometries. The numerical analysis is performed by using the boundary value problem solution, by considering separately the impedance matrix of the structure segments. The analysis for the complete structure is implemented by choosing properly the number and location of the neighboor element ports. The numerial analysis is performed for the following antenna geometries: resonant cavity, microstrip rectangular patch antenna, and microstrip rectangular patch antenna with aperture. The analysis is firstly developed for microstrip antennas on isotropic substrates, and then extended to the case of microstrip antennas on anisotropic substrates by using a Mapping Method. The experimental work is described and related to the development of several prototypes of rectangular microstrip patch antennas wtih and without rectangular apertures. A good agreement was observed between the simulated and measured results. Thereafter, a good agreement was also observed between the results of this work and those shown in literature for microstrip antennas on isotropic substrates. Furthermore, results are proposed for rectangular microstrip patch antennas wtih rectangular apertures in the conducting patch / As antenas de microfita s?o estruturas muito utilizadas nos sistemas de telecomunica??es atuais. Isto decorre, principalmente, da diversidade de configura??es e da facilidade de constru??o e integra??o dessas antenas com outros dispositivos e circuitos de altas freq??ncias. Neste trabalho, o m?todo de an?lise empregado ? o Modelo de Circuito de M?lti-Porta (Multiport Network Model MNM), que combinado com o M?todo da Segmenta??o e a t?cnica da Fun??o de Green, mostra-se adequado ao estudo da antena de microfita com abertura no patch condutor. A partir do equacionamento do problema do valor de contorno, ? ent?o realizada uma an?lise num?rica que consiste em avaliar a estrutura da antena considerada a partir da integra??o dos elementos em que ela foi dividida. Nessa an?lise, os elementos s?o representados por matrizes de imped?ncia e a integra??o ? implementada atrav?s de portas de circuitos adequadamente escolhidas em n?mero e posicionamento. Na an?lise num?rica, foram consideradas as seguintes estruturas: a cavidade ressonante, a microfita com patch retangular convencional (sem abertura) e a microfita com patch retangular com abertura. A an?lise foi efetuada para substratos isotr?picos e estendida para o caso de antenas com substratos anisotr?picos uniaxiais atrav?s do M?todo do Mapeamento. S?o apresentados resultados para a freq??ncia de resson?ncia e para a imped?ncia de entrada de antenas de microfita. A parte experimental do trabalho consistiu no projeto, constru??o e medi??o de v?rios prot?tipos de antenas de microfita com patches retangulares com e sem abertura. Observou-se que os resultados obtidos, atrav?s da simula??o num?rica, apresentaram uma boa concord?ncia com os das medi??es efetuadas. Os resultados deste trabalho, tamb?m, concordaram com os resultados de outros autores, dispon?veis na literatura

Speech segmentation is the process of identifying the boundaries between words, syllables or phones in the recorded waveforms of spoken natural languages. The lowest level of speech segmentation is the breakup and classification of the sound signal into a string of phones. The difficulty of this problem is compounded by the phenomenon of co-articulation of speech sounds. The classical solution to this problem is to manually label and segment spectrograms. In the first step of this two step process, a trained person listens to a speech signal, recognizes the word and phone sequence, and roughly determines the position of each phonetic boundary. The second step involves examining several features of the speech signal to place a boundary mark at the point where these features best satisfy a certain set of conditions specific for that kind of phonetic boundary. Manual segmentation of speech into phones is a highly time-consuming and painstaking process. Required for a variety of applications, such as acoustic analysis, or building speech synthesis databases for high-quality speech output systems, the time required to carry out this process for even relatively small speech databases can rapidly accumulate to prohibitive levels. This calls for automating the segmentation process. The state-of-art segmentation techniques use Hidden Markov Models (HMM) for phone states. They give an average accuracy of over 95% within 20 ms of manually obtained boundaries. However, HMM based methods require large training data for good performance. Another major disadvantage of such speech recognition based segmentation techniques is that they cannot handle very long utterances, Which are necessary for prosody modeling in speech synthesis applications. Development of Text to Speech (TTS) systems in Indian languages has been difficult till date owing to the non-availability of sizeable segmented speech databases of good quality. Further, no prosody models exist for most of the Indian languages. Therefore, long utterances (at the paragraph level and monologues) have been recorded, as part of this work, for creating the databases. This thesis aims at automating segmentation of very long speech sentences recorded for the application of corpus-based TTS synthesis for multiple Indian languages. In this explicit segmentation problem, we need to force align boundaries in any utterance from its known phonetic transcription. The major disadvantage of forcing boundary alignments on the entire speech waveform of a long utterance is the accumulation of boundary errors. To overcome this, we force boundaries between 2 known phones (here, 2 successive stop consonants are chosen) at a time. Here, the approach used is silence detection as a marker for stop consonants. This method gives around 89% (for Hindi database) accuracy and is language independent and training free. These stop consonants act as anchor points for the next stage. Two methods for explicit segmentation have been proposed. Both the methods rely on the accuracy of the above stop consonant detection stage. Another common stage is the recently proposed implicit method which uses Bach scale filter bank to obtain the feature vectors. The Euclidean Distance of the Mean of the Logarithm (EDML) of these feature vectors shows peaks at the point where the spectrum changes. The method performs with an accuracy of 87% within 20 ms of manually obtained boundaries and also achieves a low deletion and insertion rate of 3.2% and 21.4% respectively, for 100 sentences of Hindi database. The first method is a three stage approach. The first is the stop consonant detection stage followed by the next, which uses Quatieri’s sinusoidal model to classify sounds as voiced/unvoiced within 2 successive stop consonants. The final stage uses the EDML function of Bach scale feature vectors to further obtain boundaries within the voiced and unvoiced regions. It gives a Frame Error Rate (FER) of 26.1% for Hindi database. The second method proposed uses duration statistics of the phones of the language. It again uses the EDML function of Bach scale filter bank to obtain the peaks at the phone transitions and uses the duration statistics to assign probability to each peak being a boundary. In this method, the FER performance improves to 22.8% for the Hindi database. Both the methods are equally promising for the fact that they give low frame error rates. Results show that the second method outperforms the first, because it incorporates the knowledge of durations. For the proposed approaches to be useful, manual interventions are required at the output of each stage. However, this intervention is less tedious and reduces the time taken to segment each sentence by around 60% as compared to the time taken for manual segmentation. The approaches have been successfully tested on 3 different languages, 100 sentences each -Kannada, Tamil and English (we have used TIMIT database for validating the algorithms). In conclusion, a practical solution to the segmentation problem is proposed. Also, the algorithm being training free, language independent (ES-SABSF method) and speaker independent makes it useful in developing TTS systems for multiple languages reducing the segmentation overhead. This method is currently being used in the lab for segmenting long Kannada utterances, spoken by reading a set of 1115 phonetically rich sentences.

Bei strukturorientierten, aus relativ großen Holzpartikeln aufgebauten Holzwerkstoffen wie z.B. OSB (oriented strand board) addieren sich die gerichteten Festigkeiten der einzelnen Lagen je nach Orientierung der Partikel und der Verteilung ihrer Größenparameter. Wünschenswert wäre eine Messung der Partikelgeometrie und Orientierung möglichst im Prozess, z.B. am Formstrang vor der Presse direkt durch den „Blick auf das Vlies“. Bisher sind regelmäßige on-line-Messungen der Spangeometrie aber nicht möglich, und Einzelspanmessungen werden nicht vorgenommen, weil sie zu aufwändig wären. Um die Partikelkonturen zunächst hinreichend für die Vermessung zu restaurieren und dann zu vermessen, muss ein mehrstufiges Verfahren angewendet werden, das eine Szene mit Strands und mehr oder weniger deutlichen Kanten zunächst als „Grauwertgebirge“ auffasst. Zur Segmentierung reicht ein Watershed-Algorithmus nicht aus. Auch ein zweistufiger Kantendetektor nach Canny liefert allein noch kein ausreichendes Ergebnis, weil sich keine geschlossenen Objektkonturen ergeben. Hinreichend dagegen ist ein komplexes Verfahren auf der Grundlage der Höhenschichtzerlegung und nachfolgenden Synthese: Nach einer Transformation der Grauwerte des Bildes in eine reduzierte, gleichverteilte Anzahl von Höhenschichten werden zwischen diesen die lokalen morphologischen Gradienten berechnet und herangezogen für die Rekonstruktion der ursprünglichen Spankonturen. Diese werden aus den Höhenschichten aufaddiert, wobei allerdings nur Teilflächen innerhalb eines für die gesuchten Spangrößen plausiblen Größenintervalls einbezogen werden, um Störungen zu unterdrücken. Das Ergebnis der Rekonstruktion wird zusätzlich verknüpft mit den bereits durch einen Canny-Operator im Originalbild detektierten deutlichen Kanten und morphologisch bereinigt. Diese erweiterte Höhenschichtanalyse ergibt ausreichend segmentierte Bilder, in denen die Objektgrenzen weitgehend den Spankonturen entsprechen. Bei der nachfolgenden Vermessung der Objekte werden Standard-Algorithmen eingesetzt, wobei sich die Approximation von Spankonturen durch momentengleiche Ellipsen als sinnvoll erwies. Verbliebene Fehldetektionen können bei der Vermessung unterdrückt werden durch Formfaktoren und zusätzliche Größenintervalle. Zur Darstellung und Charakterisierung der Größenverteilungen für die Länge und die Breite wurden die nach der Objektfläche gewichtete, linear skalierte Verteilungsdichte (q2-Verteilung), die Verteilungssumme und verschiedene Quantile verwendet. Zur Umsetzung und Demonstration des Zusammenwirkens der verschiedenen Algorithmen wurde auf der Basis von MATLAB das Demonstrationsprogramm „SizeBulk“ entwickelt, das Bildfolgen verarbeiten kann und mit dem die verschiedenen Varianten der Bildaufbereitung und Parametrierung durchgespielt werden können. Das Ergebnis des Detektionsverfahrens enthält allerdings nur die vollständigen Konturen der ganz oben liegenden Objekte; Objekte unterhalb der Außenlage sind teilweise verdeckt und können daher nur unvollständig vermessen werden. Zum Test wurden daher synthetische Bilder mit vereinzelten und überlagerten Objekten bekannter Größenverteilung erzeugt und dem Detektions- und Messverfahren unterworfen. Dabei zeigte sich, dass die Größenstatistiken durch den Überlagerungseffekt und auch die Spanorientierung zwar beeinflusst werden, dass aber zumindest die Modalwerte der wichtigsten Größenparameter Länge und Breite meist erkennbar bleiben. Als Versuchsmaterial dienten außer den synthetischen Bildern verschiedene Sortimente von OSB-Strands aus Industrie- und Laborproduktion. Sie wurden sowohl manuell vereinzelt als auch zu einem Vlies arrangiert vermessen. Auch bei realen Strands zeigten sich gleiche Einflüsse der Überlagerung auf die Größenverteilungen wie in der Simulation. Es gilt aber auch hier, dass die Charakteristika verschiedener Spankontingente bei gleichen Aufnahmebedingungen und Auswerteparametern gut messbar sind bzw. dass Änderungen in der gemessenen Größenverteilung eindeutig den geometrischen Eigenschaften der Späne zugeordnet werden können. Die Eignung der Verarbeitungsfolge zur Charakterisierung von Spangrößenverteilungen bestätigte sich auch an Bildern, die ausschließlich am Vlies auf einem Formstrang aufgenommen wurden. Zusätzlich wurde nachgewiesen, dass mit der erweiterten Höhenschichtanalyse auch Bilder von Spanplattenoberflächen ausgewertet werden könnten und daraus auf die Größenverteilung der eingesetzten Deckschichtspäne geschlossen werden kann. Das vorgestellte Verfahren ist daher eine gute und neuartige Möglichkeit, prozessnah an Teilflächen von OSB-Vliesen anhand von Grauwertbildern die Größenverteilungen der Strands zu charakterisieren und eignet sich grundsätzlich für den industriellen Einsatz. Geeignete Verfahren waren zumindest für Holzpartikel bisher nicht bekannt. Diese Möglichkeit, Trends in der Spangrößenverteilung automatisch zu erkennen, eröffnet daher neue Perspektiven für die Prozessüberwachung. / The strength of wood-based materials made of several layers of big and oriented particles like OSB (oriented strand board) is a superposition of the strengths of the layers according to the orientation of the particles and depending from their size distribution. It would be desirable to measure particle geometry and orientation close to the production process, e.g. with a “view onto the mat”. Currently, continuous on-line measurements of the particle geometry are not possible, while measurements of separated particles would be too costly and time-consuming. Before measuring particle shapes they have to be reconstructed in a multi-stage procedure which considers an image scene with strands as “gray value mountains”. Segmentation using a watershed algorithm is not sufficient. Also a two-step edge detector according to Canny does not yield closed object shapes. A multi-step procedure based on threshold decomposition and recombination however is successful: The gray values in the image are transformed into a reduced and uniformly distributed set of threshold levels. The local morphological gradients between these levels are used to re-build the original particle shapes by adding the threshold levels. Only shapes with a plausible size corresponding to real particle shapes are included in order to suppress noise. The result of the reconstruction from threshold levels is then matched with the result of the strong edges in the original image, which had been detected using a Canny operator, and is finally cleaned with morphological operators. This extended threshold analysis produces sufficiently segmented images with object shapes corresponding extensively to the particle shapes. Standard algorithms are used to measure geometric features of the objects. An approximation of particle shapes with ellipses of equal moments of inertia is useful. Remaining incorrectly detected objects are removed by form factors and size intervals. Size distributions for the parameters length and width are presented and characterized as density distribution histograms, weighted by the object area and linearly scaled (q2 distribution), as well as the cumulated distribution and different quantiles. A demonstration software “SizeBulk” based on MATLAB has been developed to demonstrate the computation and the interaction of algorithms. Image sequences can be processed and different variations of image preprocessing and parametrization can be tested. However, the detection procedure yields complete shapes only for those particles in the top layer. Objects in lower layers are partially hidden and cannot be measured completely. Artificial images with separated and with overlaid objects with a known size distribution were generated to study this effect. It was shown that size distributions are influenced by this covering effect and also by the strand orientation, but that at least the modes of the most important size parameters length and width remain in evidence. Artificial images and several samples with OSB strands from industrial and laboratory production were used for testing. They were measured as single strands as well as arrangements similar to an OSB mat. For real strands, the same covering effects to the size distributions revealed as in the simulation. Under stable image acquisition conditions and using similar processing parameters the characteristics of these samples can well be measured, and changes in the size distributions are definitely due to the geometric properties of the strands. The suitability of the processing procedure for the characterization of strand size distributions could also be confirmed for images acquired from OSB mats in a production line. Moreover, it could be shown that the extended threshold analysis is also suitable to evaluate images of particle board surfaces and to draw conclusions about the size distribution of the top layer particles. Therefore, the method presented here is a novel possibility to measure size distributions of OSB strands through the evaluation of partial gray value images of the mat surface. In principle, this method is suitable to be transferred to an industrial application. So far, methods that address the problem of detecting trends of the strand size distribution were not known, and this work shows new perspectives for process monitoring.

L'analyse de la microstructure osseuse joue un rôle important pour étudier des maladies de l'os comme l'ostéoporose. Des nouveaux scanners périphériques haute résolution (HR-pQCT) permettent de faire des acquisitions de la micro-architecture osseuse in-vivo sur l'homme. Toutefois la résolution spatiale de ces appareils reste comparable à la taille des travées osseuses, ce qui limite leur analyse quantitative. L'objectif de cette thèse est de proposer de nouvelles approches jointes super-résolution/ segmentation pour une analyse quantitative plus fine des images HR-pQCT in-vivo de la structure osseuse trabéculaire. Dans une première étape nous nous sommes concentrés sur des méthodes 2D de super-résolution avec régularisation par variation totale (TV) puis par variation totale d'ordre plus élevé (Higher Degree TV), avec minimisation par un algorithme ADMM (Alternating Direction Method of Multipliers). Ensuite, nous avons proposé une méthode itérative combinant le principe de Morozov et la méthode de Newton pour estimer le paramètre de régularisation TV. Comparé à la méthode UPRE (Unbiased Predictive Risk Estimator), la méthode proposée est plus rapide et ne requiert pas un balayage exhaustif des valeurs des paramètres. Nous avons développé dans une deuxième étape une méthode de super-résolution/segmentation conjointe avec un a priori basé sur la Variation Totale et une relaxation convexe (Tvbox), qui permet d'améliorer les paramètres quantitatifs de l'os et de la connectivité 3D. La méthode a été validée sur des images expérimentales micro-CT déteriorées artificiellement. Finalement, en vue de l'application à des images réelles HR-pQCT, nous nous sommes intéressés à une approche conjointe semi-aveugle super-résolution/segmentation qui vise à estimer à la fois l'image binaire super-résolue et le noyau de convolution. Des résultats sur des images micro-CT et HR-pQCT sont présentés. En conclusion, notre travail montre que les méthodes d'optimisation basées sur la régularisation TV sont prometteurs pour améliorer la quantification de la micro-architecture osseuse sur des images HR-pQCT. / The investigation of trabecular bone micro-architecture provides relevant information to determine the bone strength, an important parameter in osteoporosis investigation. While the spatial resolution of clinical CT is not sufficient to resolve the trabecular structure, the High Resolution peripheral Quantitative CT (HR-pQCT) has been developed to investigate bone micro-architecture in-vivo at peripheral sites (tibia and radius). Despite this considerable progress, the quantification of 3D trabecular bone micro-architecture in-vivo remains limited due to a lack of spatial resolution compared to the trabeculae size. The objective of this thesis is to propose new joint super-resolution/segmentation approaches for improving the quantitative analysis of in-vivo HR-pQCT images of the trabecular bone structure. To begin with, we have investigated 2D super-resolution methods based on Total Variation (TV) and Higher Degree Total Variation (HDTV) and Alternating Direction Method of Multipliers (ADMM) minimization. Afterwards, an iterative method combining the Morozov principle and the Newton method was proposed in order to estimate the TV regularization parameter. The proposed method provides a very good regularization parameter only in few iterations compared with the UPRE method that requires an extensive scanning of parameter values. Furthermore, we have developed a 3D joint super-resolution/segmentation method based on a TV a prior with a convex relaxation (TVbox). The validation of the proposed methods was made on experimental micro-CT bone images artificially deteriorated. The results showed an improvement of the bone parameters and 3D connectivity with the TVbox method. Moreover, we have investigated a semi-blind joint super-resolution/ segmentation approach aiming to estimate both the binary super-resolved image and the assumed Gaussian blurring kernel that is not known for the real HR-pQCT images. Results on micro-CT and HR-pQCT experimental bone images were presented. In conclusion, our work has shown that TV based regularization methods promise to improve the quantification of bone micro-architecture from HR-pQCT images.

Bei strukturorientierten, aus relativ großen Holzpartikeln aufgebauten Holzwerkstoffen wie z.B. OSB (oriented strand board) addieren sich die gerichteten Festigkeiten der einzelnen Lagen je nach Orientierung der Partikel und der Verteilung ihrer Größenparameter. Wünschenswert wäre eine Messung der Partikelgeometrie und Orientierung möglichst im Prozess, z.B. am Formstrang vor der Presse direkt durch den „Blick auf das Vlies“. Bisher sind regelmäßige on-line-Messungen der Spangeometrie aber nicht möglich, und Einzelspanmessungen werden nicht vorgenommen, weil sie zu aufwändig wären. Um die Partikelkonturen zunächst hinreichend für die Vermessung zu restaurieren und dann zu vermessen, muss ein mehrstufiges Verfahren angewendet werden, das eine Szene mit Strands und mehr oder weniger deutlichen Kanten zunächst als „Grauwertgebirge“ auffasst. Zur Segmentierung reicht ein Watershed-Algorithmus nicht aus. Auch ein zweistufiger Kantendetektor nach Canny liefert allein noch kein ausreichendes Ergebnis, weil sich keine geschlossenen Objektkonturen ergeben. Hinreichend dagegen ist ein komplexes Verfahren auf der Grundlage der Höhenschichtzerlegung und nachfolgenden Synthese: Nach einer Transformation der Grauwerte des Bildes in eine reduzierte, gleichverteilte Anzahl von Höhenschichten werden zwischen diesen die lokalen morphologischen Gradienten berechnet und herangezogen für die Rekonstruktion der ursprünglichen Spankonturen. Diese werden aus den Höhenschichten aufaddiert, wobei allerdings nur Teilflächen innerhalb eines für die gesuchten Spangrößen plausiblen Größenintervalls einbezogen werden, um Störungen zu unterdrücken. Das Ergebnis der Rekonstruktion wird zusätzlich verknüpft mit den bereits durch einen Canny-Operator im Originalbild detektierten deutlichen Kanten und morphologisch bereinigt. Diese erweiterte Höhenschichtanalyse ergibt ausreichend segmentierte Bilder, in denen die Objektgrenzen weitgehend den Spankonturen entsprechen. Bei der nachfolgenden Vermessung der Objekte werden Standard-Algorithmen eingesetzt, wobei sich die Approximation von Spankonturen durch momentengleiche Ellipsen als sinnvoll erwies. Verbliebene Fehldetektionen können bei der Vermessung unterdrückt werden durch Formfaktoren und zusätzliche Größenintervalle. Zur Darstellung und Charakterisierung der Größenverteilungen für die Länge und die Breite wurden die nach der Objektfläche gewichtete, linear skalierte Verteilungsdichte (q2-Verteilung), die Verteilungssumme und verschiedene Quantile verwendet. Zur Umsetzung und Demonstration des Zusammenwirkens der verschiedenen Algorithmen wurde auf der Basis von MATLAB das Demonstrationsprogramm „SizeBulk“ entwickelt, das Bildfolgen verarbeiten kann und mit dem die verschiedenen Varianten der Bildaufbereitung und Parametrierung durchgespielt werden können. Das Ergebnis des Detektionsverfahrens enthält allerdings nur die vollständigen Konturen der ganz oben liegenden Objekte; Objekte unterhalb der Außenlage sind teilweise verdeckt und können daher nur unvollständig vermessen werden. Zum Test wurden daher synthetische Bilder mit vereinzelten und überlagerten Objekten bekannter Größenverteilung erzeugt und dem Detektions- und Messverfahren unterworfen. Dabei zeigte sich, dass die Größenstatistiken durch den Überlagerungseffekt und auch die Spanorientierung zwar beeinflusst werden, dass aber zumindest die Modalwerte der wichtigsten Größenparameter Länge und Breite meist erkennbar bleiben. Als Versuchsmaterial dienten außer den synthetischen Bildern verschiedene Sortimente von OSB-Strands aus Industrie- und Laborproduktion. Sie wurden sowohl manuell vereinzelt als auch zu einem Vlies arrangiert vermessen. Auch bei realen Strands zeigten sich gleiche Einflüsse der Überlagerung auf die Größenverteilungen wie in der Simulation. Es gilt aber auch hier, dass die Charakteristika verschiedener Spankontingente bei gleichen Aufnahmebedingungen und Auswerteparametern gut messbar sind bzw. dass Änderungen in der gemessenen Größenverteilung eindeutig den geometrischen Eigenschaften der Späne zugeordnet werden können. Die Eignung der Verarbeitungsfolge zur Charakterisierung von Spangrößenverteilungen bestätigte sich auch an Bildern, die ausschließlich am Vlies auf einem Formstrang aufgenommen wurden. Zusätzlich wurde nachgewiesen, dass mit der erweiterten Höhenschichtanalyse auch Bilder von Spanplattenoberflächen ausgewertet werden könnten und daraus auf die Größenverteilung der eingesetzten Deckschichtspäne geschlossen werden kann. Das vorgestellte Verfahren ist daher eine gute und neuartige Möglichkeit, prozessnah an Teilflächen von OSB-Vliesen anhand von Grauwertbildern die Größenverteilungen der Strands zu charakterisieren und eignet sich grundsätzlich für den industriellen Einsatz. Geeignete Verfahren waren zumindest für Holzpartikel bisher nicht bekannt. Diese Möglichkeit, Trends in der Spangrößenverteilung automatisch zu erkennen, eröffnet daher neue Perspektiven für die Prozessüberwachung. / The strength of wood-based materials made of several layers of big and oriented particles like OSB (oriented strand board) is a superposition of the strengths of the layers according to the orientation of the particles and depending from their size distribution. It would be desirable to measure particle geometry and orientation close to the production process, e.g. with a “view onto the mat”. Currently, continuous on-line measurements of the particle geometry are not possible, while measurements of separated particles would be too costly and time-consuming. Before measuring particle shapes they have to be reconstructed in a multi-stage procedure which considers an image scene with strands as “gray value mountains”. Segmentation using a watershed algorithm is not sufficient. Also a two-step edge detector according to Canny does not yield closed object shapes. A multi-step procedure based on threshold decomposition and recombination however is successful: The gray values in the image are transformed into a reduced and uniformly distributed set of threshold levels. The local morphological gradients between these levels are used to re-build the original particle shapes by adding the threshold levels. Only shapes with a plausible size corresponding to real particle shapes are included in order to suppress noise. The result of the reconstruction from threshold levels is then matched with the result of the strong edges in the original image, which had been detected using a Canny operator, and is finally cleaned with morphological operators. This extended threshold analysis produces sufficiently segmented images with object shapes corresponding extensively to the particle shapes. Standard algorithms are used to measure geometric features of the objects. An approximation of particle shapes with ellipses of equal moments of inertia is useful. Remaining incorrectly detected objects are removed by form factors and size intervals. Size distributions for the parameters length and width are presented and characterized as density distribution histograms, weighted by the object area and linearly scaled (q2 distribution), as well as the cumulated distribution and different quantiles. A demonstration software “SizeBulk” based on MATLAB has been developed to demonstrate the computation and the interaction of algorithms. Image sequences can be processed and different variations of image preprocessing and parametrization can be tested. However, the detection procedure yields complete shapes only for those particles in the top layer. Objects in lower layers are partially hidden and cannot be measured completely. Artificial images with separated and with overlaid objects with a known size distribution were generated to study this effect. It was shown that size distributions are influenced by this covering effect and also by the strand orientation, but that at least the modes of the most important size parameters length and width remain in evidence. Artificial images and several samples with OSB strands from industrial and laboratory production were used for testing. They were measured as single strands as well as arrangements similar to an OSB mat. For real strands, the same covering effects to the size distributions revealed as in the simulation. Under stable image acquisition conditions and using similar processing parameters the characteristics of these samples can well be measured, and changes in the size distributions are definitely due to the geometric properties of the strands. The suitability of the processing procedure for the characterization of strand size distributions could also be confirmed for images acquired from OSB mats in a production line. Moreover, it could be shown that the extended threshold analysis is also suitable to evaluate images of particle board surfaces and to draw conclusions about the size distribution of the top layer particles. Therefore, the method presented here is a novel possibility to measure size distributions of OSB strands through the evaluation of partial gray value images of the mat surface. In principle, this method is suitable to be transferred to an industrial application. So far, methods that address the problem of detecting trends of the strand size distribution were not known, and this work shows new perspectives for process monitoring.