Topics in image processing and applications to structural biology and object segmentation
Federico Lecumberry
PhD thesis from Universidad de la República (Uruguay). Facultad de Ingeniería. IIE - Mar. 2012
Advisor: Guillermo Sapiro
Co-advisor: Alvaro Pardo
Research Group(s): Tratamiento de Imagenes (gti)
Department(s): Procesamiento de Señales
Descargar la publicación : Lec12.pdf [20.7Mo]  


This thesis is divided in two parts, addressing quite different topics and fields such as Cryo-Electron Microscopy and Object Segmentation. Cryo-Electron Microscopy The first part addresses the problem of 3D reconstruction of macromolecular assemblies from their 2D random projection images obtained following a tomographic acquisition in a Transmission Electron Microscope. The challenge of this problem comes from the extremely noisy and low contrasted projection images obtained with this procedure. The problem of reconstruction from random projections or multiple view geometry is well known in Computer Vision, however the usual techniques based in identifying correspondent point or lines in several views are discarded, due to the noisy characteristics of the images (see for example figure 2.1b in page 7). Thus, usually thousands of aligned projections are averaged in order to increase the SNR and compute the reconstruction. The main reason for the low SNR is the low electron doses allowed to irradiate the macromolecules of interest, otherwise severe radiation damage accumulates, deforming the macromolecules and affecting the projections. Embedding the specimens in vitreous ice and preserving it at cryogenic temperatures (below -150◦ Celsius) helps reduce the accumulated radiation damage. In this scenario, Cryo-Electron Microscopy has proved to be a powerful technique to obtain 3D reconstructions with a wide range of resolutions. Two of the most popular techniques in Cryo-Electron Microscopy are Single Particle Analysis and Cryo-Electron Tomography, and are described in this work. Their main differences for the goal of this work are that the former is capable to routinely obtain reconstruction ∼8Å or better of in vitro specimens, and the later allows to analyze in vivo specimens but hardly recovers reconstructions better than 20Å. This work proposes a new framework combining the Single Particle Analysis and Cryo-Electron Tomography approaches, that combines the best of both worlds, adding the high resolution feature in the maps obtained with the CryoElectron Tomography data collection procedure without sacrificing its desired features. This is achieved through a new refinement algorithm and an innovative adaptation of the Single Particle Analysis reconstruction procedure to this kind of data. The framework is validated by a set of experiments, first, synthesizing the data collection procedure, allowing to have access to the complete groundtruth (parameters and macromolecule structure) for the comparison. The second set of experiments use data acquired in the Transmission Electron Microscope and only the macromolecule structure is known. The reconstructed density map for an homogeneous macromolecule shows details of about 10Å of resolutions. This work was performed in collaboration with Alberto Bartesaghi and Sriram Subramaniam from the Laboratory of Cell Biology, Center for Cancer Research, National Cancer Institute, National Institutes of Health, USA. This manuscript was prepared before the publication of this work, further information can be found in my web page: ~fefo/. Object segmentation. The second part addresses one of the most interesting and fundamental tasks in Computer Vision and Image Processing such as object segmentation. In this work we use the idea of Shape Models (SMs). SMs, capturing the common features of a set of training shapes, represent a new incoming object based on its projection onto the corresponding model. Given a set of learned SMs representing different objects classes, and an image with a new shape, this work introduces a joint classification-segmentation framework with a twofold goal. First, to automatically select the SM that best represents the object, and second, to accurately segment the image taking into account both the image information and the features and variations learned from the on-line selected model. A new energy functional is introduced that simultaneously accomplishes both goals. Position and transformation invariance is included as part of the modeling as well. The model selection is performed based on a shape similarity measure, online determining which model to use at each iteration of the steepest descent minimization, allowing for model switching and adaptation to the data. Highorder SMs are used in order to deal with very similar object classes and natural variability within them. The presentation of the framework is complemented with examples for the difficult task of simultaneously classifying and segmenting closely related shapes, such as stages of human activities, in images with severe occlusions. This work was presented in the IEEE International Conference on Image Processing, ICIP 2009 [1], and appears in the IEEE Transactions on Image Processing [2]. The contents of the second part is almost the same that appears in the later reference

Datos adicionales


Referencias BibTex

Descargar BibTex bibtex

Otras publicaciones

» Federico Lecumberry