A segmentação de imagens consiste em urna tarefa de fundamental importância para diferentes aplicações em visão computacional, tais como por exemplo, o reconhecimento e o rastreamento de objetos, a segmentação de tomores/lesões em aplicações médicas, podendo também servir de auxílio em sistemas de reconhecimento facial. Embora exista uma extensa literatora abordando o problema de segmentação de imagens, tal tópico ainda continua em aberto para pesquisa. Particularmente, a tarefa de segmentar imagens coloridas é desafiadora devido as diversas inomogeneidades de cor, texturas e formas presentes nas feições descritivas das imagens. Este trabalho apresenta um novo método de clustering para abordar o problema da segmentação de imagens coloridas. Nós desenvolvemos uma abordagem Bayesiana para procura de máximos de densidade em urna distribuição discreta de dados, e representamos os dados de forma hierárquica originando clusters adaptativos a cada nível da hierarquia. Nós aplicamos o método de clustering proposto no problema de segmentação de imagens coloridas, aproveitando sua estrutura hierárquica, baseada em propriedades de árvores direcionadas, para representar hierarquicamente uma imagem colorida. Os experimentos realizados revelaram que o método de clustering proposto, aplicado ao problema de segmentação de imagens coloridas, obteve para a medida de performance Probabilistic Rand lndex (PRI) o valor de 0.8148 e para a medida Global Consistency Error (GCE) o valor 0.1701, superando um total de vinte e um métodos previamente propostos na literatura para o banco de dados BSD300. Comparações visuais confirmaram a competitividade da nossa abordagem em relação aos demais métodos testados. Estes resultados enfatizam a potencialidade do nosso método de clustering para abordar outras aplicações no domínio de Visão Computacional e Reconhecimento de Padrões. / Image segmentation is an essential task for several computer vision applications, such as object recognition, tracking and image retrieval. Although extensively studied in the literature, the problem of image segmentation remains an open topic of research. Particularly, the task of segmenting color images is challenging due to the inhomogeneities in the color regions encountered in natural scenes, often caused by the shapes of surfaces and their interactions with the illumination sources (e.g. causing shading and highlights) This work presents a novel non-supervised classification method. We develop a Bayesian framework for seeking modes on the underlying discrete distribution of data and we represent data hierarchically originating adaptive clusters at each levei of hierarchy. We apply the prnposal clustering technique for tackling the problem of color irnage segmentation, taking advantage of its hierarchical structure based on hierarchy properties of directed trees for representing fine to coarse leveis of details in an image. The experiments herein conducted revealed that the proposed clustering method applied to the color image segmentation problem, achieved for the Probabilistic Rand Index (PRI) performance measure the value of 0.8148 and for the Global Consistency Error (GCE) the value of 0.1701, outperforming twenty-three methods previously proposed in the literature for the BSD300 dataset. Visual comparison confirmed the competitiveness of our approach towards state-of-art methods publicly available in the literature. These results emphasize the great potential of our proposed clustering technique for tackling other applications in computer vision and pattem recognition.

Segmentação de imagens é ainda uma etapa desafiadora do processo de reconhecimento de padrões. Entre as abordagens de segmentação, muitas são baseadas em particionamento em grafos, as quais apresentam alguns inconvenientes, sendo um deles o tempo de processamento muito elevado. Com as recentes pesquisas na teoria de redes complexas, as técnicas de reconhecimento de padrões baseadas em grafos melhoraram consideravelmente. A identificação de grupos de vértices pode ser considerada um processo de detecção de comunidades de acordo com a teoria de redes complexas. Como o agrupamento de dados está relacionado com a segmentação de imagens, esta também pode ser abordada através de redes complexas. No entanto, a segmentação de imagens baseado em redes complexas apresenta uma limitação fundamental, que é o número excessivo de nós na rede. Neste trabalho é proposta uma abordagem de redes complexas para segmentação de imagens de grandes dimensões que é ao mesmo tempo precisa e rápida. Para alcançar este objetivo, é incorporado o conceito de Super Pixels, visando reduzir o número de nós da rede. Os experimentos mostraram que a abordagem proposta produz segmentações de boa qualidade em baixo tempo de processamento. Além disso uma das principais contribuições deste trabalho é a determinação dos melhores parâmetros, uma vez que torna o método bastante independente dos parâmetros, o que não fora alcançado antes em nenhuma pesquisa da área / Image segmentation is still a challenging stage of the pattern recognition process. Amongst the various segmentation approaches, some are based on graph partitioning, many of which show some drawbacks, such as the high processing times. Recent trends on complex network theory have contributed considerably to the development of graph-based pattern recognition techniques. The identification of group of vertices can be considered a community detection process according to complex network theory. Since data clustering is closely related to image segmentation, image segmentation tasks can also be tackled by complex networks. However, complex network-based image segmentation poses a very important limitation: the excessive number of nodes of the underlying network. In this work we propose a approach based on complex networks suitable for the segmentation of image with large dimensions that is accurate and yet fast. To accomplish that, we have incorporated the concept of Super Pixels aiming at reducing the number of the nodes in the network. The results have shown that the proposed approach delivered accurate image segmentation within low computational times. Another contribution worth mentioning is the determination of the best values for the parameters needed by the underlying graphbased segmentation and community detection algorithms, which enabled the proposed approach to become less dependent on the parameters. To the best of our knowledge, this is a new contribution to the field

Knee osteoarthritis is one of the most debilitating aging diseases as it causes loss of cartilage of the knee joint. Knee osteoarthritis affects the quality of life and increases the burden on health care costs. With no disease-modifying osteoarthritis drug currently available there is an immediate need to understand the factors triggering the onset and progression of the disease. Developing robust segmentation techniques and quantitative analysis helps identify potential imaging-based biomarkers that indicate the onset and progression of osteoarthritis. This thesis work developed layered optimal graph image segmentation of multiple objects and surfaces (LOGISMOS) framework based knee MRI segmentation algorithms in 3D and longitudinal 3D (4D). A hierarchical random forest classifier algorithm was developed to improve cartilage costs functions for the LOGISMOS framework. The new cost function design significantly improved the segmentation accuracy over the existing state of the art methods. Disease progression results in more artifacts appearing similar to cartilage in MRI. 4D LOGISMOS segmentation was developed to simultaneously segment multiple time-points of a single patient by incorporating information from earlier time points with a relatively healthier knee in the early stage of the disease. Our experiments showed consistently higher segmentation accuracy across all the time-points over 3D LOGISMOS segmentation of each time-point. Fully automated segmentation algorithms proposed are not 100% accurate especially for patient MRI's having severe osteoarthritis and require interactive correction. An interactive technique called just-enough interaction (JEI) was developed which added a fast correction step to the automated LOGISMOS, speeding up the interactions substantially over the current slice-by-slice manual editing while maintaining high accuracy. JEI editing modifies the graph nodes instead of the boundary surfaces of the bones and cartilages providing globally optimally corrected results. 3D JEI was extended to 4D JEI allowing for simultaneous visualization and interaction of multiple time points of the same patients. Further quantitative analysis tools were developed to study the thickness losses. Nomenclature compliant sub-plate detection algorithm was developed to quantify thickness in the smaller load bearing regions of the knee to help understand the varying rates of thickness losses in the different regions. Regression models were developed to predict the thickness accuracy on a patient MRI at a later follow-up using the available thickness information from the LOGISMOS segmentation of the current set of MRI scans of the patient. Further non-cartilage based imaging biomarker quantification was developed to analyze bone shape changes between progressing and non-progressing osteoarthritic populations. The algorithm quantified statistically significant local shape changes between the two populations. Overall this work improved the state of the art in the segmentation of the bones and cartilage of the femur and tibia. Interactive 3D and 4D JEI were developed allowing for fast corrections of the segmentations and thus significantly improving the accuracy while performing many times faster. Further, the quantitative analysis tools developed robustly analyzed the segmentation providing measurable metrics of osteoarthritis progression.

Convolutional neural networks (ConvNets) have quickly become the most widely used tool for image perception and interpretation tasks over the past several years. The single most important resource needed for training a ConvNet that will successfully generalize to unseen examples is an adequately sized labeled dataset. In many interesting medical imaging cases, the necessary size or quality of training data is not suitable for directly training a ConvNet. Furthermore, access to the expertise to manually label such datasets is often infeasible. To address these barriers, we investigate a method for iterative refinement of the ConvNet training. Initially, unlabeled images are attained, minimal labeling is performed, and a model is trained on the sparse manual labels. At the end of each training iteration, full images are predicted, and additional manual labels are identified to improve the training dataset. In this work, we show how to utilize patch-based ConvNets to iteratively build a training dataset for automatically segmenting MRI images of the human cerebellum. We construct this training dataset using a small collection of high-resolution 3D images and transfer the resulting model to a much larger, much lower resolution, collection of images. Both T1-weighted and T2-weighted MRI modalities are utilized to capture the additional features that arise from the differences in contrast between modalities. The objective is to perform tissue-level segmentation, classifying each volumetric pixel (voxel) in an image as white matter, gray matter, or cerebrospinal fluid (CSF). We will present performance results on the lower resolution dataset, and report achieving a 12.7% improvement in accuracy over the existing segmentation method, expectation maximization. Further, we will present example segmentations from our iterative approach that demonstrate it’s ability to detect white matter branching near the outer regions of the anatomy, which agrees with the known biological structure of the cerebellum and has typically eluded traditional segmentation algorithms.

<p>Image segmentation is the problem of partitioning an image into meaningful parts, often consisting of an object and background. As an important part of many imaging applications, e.g. face recognition, tracking of moving cars and people etc, it is of general interest to design robust and fast segmentation algorithms. However, it is well accepted that there is no general method for solving all segmentation problems. Instead, the algorithms have to be highly adapted to the application in order to achieve good performance. In this thesis, we will study segmentation methods for blood vessels in medical images. The need for accurate segmentation tools in medical applications is driven by the increased capacity of the imaging devices. Common modalities such as CT and MRI generate images which simply cannot be examined manually, due to high resolutions and a large number of image slices. Furthermore, it is very difficult to visualize complex structures in three-dimensional image volumes without cutting away large portions of, perhaps important, data. Tools, such as segmentation, can aid the medical staff in browsing through such large images by highlighting objects of particular importance. In addition, segmentation in particular can output models of organs, tumors, and other structures for further analysis, quantification or simulation.</p><p>We have divided the segmentation of blood vessels into two parts. First, we model the vessels as a collection of lines and edges (linear structures) and use filtering techniques to detect such structures in an image. Second, the output from this filtering is used as input for segmentation tools. Our contributions mainly lie in the design of a multi-scale filtering and integration scheme for de- tecting vessels of varying widths and the modification of optimization schemes for finding better segmentations than traditional methods do. We validate our ideas on synthetical images mimicking typical blood vessel structures, and show proof-of-concept results on real medical images.</p>

Forest ecosystems have recently received worldwide attention due to their biological diversity and their major role in the global carbon balance. Detecting forest cover change is crucial for reporting forest status and assessing the evolution of forested areas. However, existing change detection approaches based on satellite remote sensing are not quite appropriate to rapidly process the large volume of earth observation data. Recent advances in image segmentation have led to new opportunities for a new object-based monitoring system. <br> <br> This thesis aims at developing and evaluating an automated object-based change detection method dedicated to high spatial resolution satellite images for identifying and mapping forest cover changes in different ecosystems. This research characterized the spectral reflectance dynamics of temperate forest stand cycle and found the use of several spectral bands better for the detection of forest cover changes than with any single band or vegetation index over different time periods. Combining multi-date image segmentation, image differencing and a dedicated statistical procedure of multivariate iterative trimming, an automated change detection algorithm was developed. This process has been further generalized in order to automatically derive an up-to-date forest mask and detect various deforestation patterns in tropical environment.<br> <br> Forest cover changes were detected with very high performances (>90 %) using 3 SPOT-HRVIR images over temperate forests. Furthermore, the overall results were better than for a pixel-based method. Overall accuracies ranging from 79 to 87% were achieved using SPOT-HRVIR and Landsat ETM imagery for identifying deforestation for two different case studies in the Virunga National Park (DRCongo). Last but not least, a new multi-scale mapping solution has been designed to represent change processes using spatially-explicit maps, i.e. deforestation rate maps. By successfully applying these complementary conceptual developments, a significant step has been done toward an operational system for monitoring forest in various ecosystems.

sA very fruitful approach to the solution of image segmentation andssurface reconstruction tasks is their formulation as estimationsproblems via the use of Markov random field models and Bayes theory.sHowever, the Maximuma Posteriori (MAP) estimate, which is the one mostsfrequently used, is suboptimal in these cases. We show that forssegmentation problems the optimal Bayesian estimator is the maximizersof the posterior marginals, while for reconstruction tasks, thesthreshold posterior mean has the best possible performance. We presentsefficient distributed algorithms for approximating these estimates insthe general case. Based on these results, we develop a maximumslikelihood that leads to a parameter-free distributed algorithm forsrestoring piecewise constant images. To illustrate these ideas, thesreconstruction of binary patterns is discussed in detail.

This dissertation tries to solve the place recognition and loop closing problem in a way similar to human visual system. First, a novel image segmentation algorithm is developed. The image segmentation algorithm is based on a Perceptual Organization model, which allows the image segmentation algorithm to ‘perceive’ the special structural relations among the constituent parts of an unknown object and hence to group them together without object-specific knowledge. Then a new object recognition method is developed. Based on the fairly accurate segmentations generated by the image segmentation algorithm, an informative object description that includes not only the appearance (colors and textures), but also the parts layout and shape information is built. Then a novel feature selection algorithm is developed. The feature selection method can select a subset of features that best describes the characteristics of an object class. Classifiers trained with the selected features can classify objects with high accuracy. In next step, a subset of the salient objects in a scene is selected as landmark objects to label the place. The landmark objects are highly distinctive and widely visible. Each landmark object is represented by a list of SIFT descriptors extracted from the object surface. This object representation allows us to reliably recognize an object under certain viewpoint changes. To achieve efficient scene-matching, an indexing structure is developed. Both texture feature and color feature of objects are used as indexing features. The texture feature and the color feature are viewpoint-invariant and hence can be used to effectively find the candidate objects with similar surface characteristics to a query object. Experimental results show that the object-based place recognition and loop detection method can efficiently recognize a place in a large complex outdoor environment.

A great number of methods based on local features or global appearances have been proposed in the literature for geospatial object detection and recognition from satellite images. However, since these approaches do not have enough discriminative capabilities between object and non-object classes, they produce results with innumerable false positives during their detection process. Moreover, due to the sliding window mechanisms, these algorithms cannot yield exact location information for the detected objects. Therefore, a geospatial object recognition algorithm based on the object shape mask is proposed to minimize the aforementioned imperfections. In order to develop such a robust recognition system, foreground extraction performance of some of popular fully and semi-automatic image segmentation algorithms, such as normalized cut, k-means clustering, mean-shift for fully automatic, and interactive Graph-cut, GrowCut, GrabCut for semi-automatic, are evaluated in terms of their subjective and objective qualities. After this evaluation, the retrieval performance of some shape description techniques, such as ART, Hu moments and Fourier descriptors, are investigated quantitatively. In the proposed system, first of all, some hypothesis points are generated for a given test image. Then, the foreground extraction operation is achieved via GrabCut algorithm after utilizing these hypothesis points as if these are user inputs. Next, the extracted binary object masks are described by means of the integrated versions of shape description techniques. Afterwards, SVM classifier is used to identify the target objects. Finally, elimination of the multiple detections coming from the generation of hypothesis points is performed by some simple post-processing on the resultant masks. Experimental results reveal that the proposed algorithm has promising results in terms of accuracy in recognizing many geospatial objects, such as airplane and ship, from high resolution satellite imagery.

The haptic exploration of 2-D images is a challenging problem in computer haptics. Research on the topic has primarily been focused on the exploration of maps and curves. This thesis describes the design and implementation of a system for the haptic exploration of photographs. The system builds on various research directions related to assistive technology, computer haptics, and image segmentation. An object-level segmentation hierarchy is generated from the source photograph to be rendered haptically as a contour image at multiple levels-of-detail. A tool for the authoring of object-level hierarchies was developed, as well as an innovative type of user interaction by region selection for accurate and efficient image segmentation. According to an objective benchmark measuring how the new method compares with other interactive image segmentation algorithms shows that our region selection interaction is a viable alternative to marker-based interaction. The hierarchy authoring tool combined with precise algorithms for image segmentation can build contour images of the quality necessary for the images to be understood by touch with our system. The system was evaluated with a user study of 24 sighted participants divided in different groups. The first part of the study had participants explore images using haptics and answer questions about them. The second part of the study asked the participants to identify images visually after haptic exploration. Results show that using a segmentation hierarchy supporting multiple levels-of-detail of the same image is beneficial to haptic exploration. As the system gains maturity, it is our goal to make it available to blind users.