Orientador: Alexandre Xavier Falcão / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-27T03:30:14Z (GMT). No. of bitstreams: 1 PhellanAro_Renzo_M.pdf: 4734368 bytes, checksum: 27f258762d497b786df234144140e47a (MD5) Previous issue date: 2014 / Resumo: A segmentação de imagens médicas consiste de duas tarefas fortemente acopladas: reconhecimento e delineamento. O reconhecimento indica a localização aproximada de um objeto, enquanto o delineamento define com precisão sua extensão espacial na imagem. O reconhecimento também verifica a corretude do delineamento do objeto. Os seres humanos são superiores aos computadores na tarefa de reconhecimento, enquanto o contrário acontece no delineamento. A segmentação manual, por exemplo, é geralmente passível de erro, tediosa, demorada e sujeita à variabilidade. Portanto, os métodos de segmentação interativa mais eficaces limitam a intervenção humana ao reconhecimento. No caso das imagens médicas, os objetos podem ser as estruturas anatômicas do corpo humano, como órgãos, sistemas e tumores. Sua segmentação é uma fase fundamental para obter medidas, como seus tamanhos e distâncias, para poder realizar sua análise quantitativa. A visualização de suas formas em 3D também é importante para sua análise qualitativa. Ambas análises podem ajudar os especialistas a estudar os fenómenos anatômicos e fisiológicos do corpo humano, diferenciar situações normais e anormais, diagnosticar doenças, estabelecer tratamentos, monitorar a evolução dos tumores e planejar procedimentos cirúrgicos. No entanto, um desafio crucial para a segmentação automática é obter um modelo matemático que possa substituir os seres humanos, capaz de reconhecer as estruturas anatômicas com base em suas características de textura e forma. Esta dissertação estuda duas aproximações importantes para este problema: os Modelos Estatísticos da Forma do Objeto (SOSMs) e os Modelos Nebulosos da Forma do Objeto (FOSMs). Os SOSMs são popularmente conhecidos como métodos de segmentação baseados em atlas e têm sido utilizados amplamente e com suceso em muitas aplicações. Porém, eles precisam do registro deformável das imagens --- um processo demorado que mapeia as imagens em um mesmo sistema de coordenadas (referência), que limita seu uso em estudos com grandes conjuntos de imagens. Os FOSMs são modelos mais recentes que podem ser significativamente mais eficientes que os SOSMs, mas precisam de métodos mais eficazes de reconhecimento e delineamento. Esta dissertação compara pela primeira vez os prós e contras dos SOSMs e FOSMs, utilizando conjuntos de imagens médicas de diferentes modalidades e estruturas anatômicas / Abstract: Image segmentation consists of two tightly coupled tasks: recognition and delineation. Recognition indicates the whereabouts of a desired object, while delineation precisely defines its spatial extent in the image. Recognition also verifies the correctness of the object's delineation. Humans are superior to computers in recognition and the other way around is valid for delineation. Manual segmentation, for instance, is usually considered error-prone, tedious, time-consuming, and subject to inter-observer variability. Therefore, the most effective interactive segmentation methods reduce human intervention to the recognition tasks. In medical images, objects may be body anatomical structures, such as organs, organ systems, and tumors. Their segmentation is a fundamental step to extract measures, such as sizes and distances for quantitative analysis. The visualization of their 3D shapes is also important for qualitative analysis. Both can help experts to study anatomical and physiological phenomena of the human body, differentiate between normal and abnormal, diagnose a disease, establish a treatment, monitor the evolution of a tumor, and plan a surgical procedure. However, a crucial challenge in automated segmentation is to obtain a surrogate mathematical model for humans, able to recognize the anatomy of such structures based on their texture and shape properties. This dissertation investigates two important approaches for this problem: the Statistical Object Shape Models (SOSMs) and the Fuzzy Object Shape Models (FOSMs). SOSMs are popularly known as atlas-based segmentation methods and have been extensively and successfully used in many applications. However, they require deformable image registration --- a time-consuming operation to map images into a common (reference) coordinate system, which limits their use in studies with large image datasets. FOSMs are more recent and can be significantly more efficient than SOSMs, but they require more effective recognition and delineation methods. This dissertation compares for the first time the pros and cons of SOSMs and FOSMs, using image datasets from distinct medical imaging modalities and anatomical structures of the human body / Mestrado / Ciência da Computação / Mestre em Ciência da Computação

Segmenting brain MR scans could be highly benecial for diagnosing, treating and evaluating the progress of specic diseases. Up to this point, manual segmentation,performed by experts, is the conventional method in hospitals and clinical environments. Although manual segmentation is accurate, it is time consuming, expensive and might not be reliable. Many non-automatic and semi automatic methods have been proposed in the literature in order to segment MR brain images, but the levelof accuracy is not comparable with manual segmentation. The aim of this project is to implement and make a preliminary evaluation of a method based on machine learning technique for segmenting gray matter (GM),white matter (WM) and cerebrospinal uid (CSF) of brain MR scans using images available within the open MICCAI grand challenge (MRBrainS13).The proposed method employs supervised articial neural network based autocontext algorithm, exploiting intensity-based, spatial-based and shape model-basedlevel set segmentation results as features of the network. The obtained average results based on Dice similarity index were 97.73%, 95.37%, 82.76%, 88.47% and 84.78% for intracranial volume, brain (WM + GM), CSF, WM and GM respectively. This method achieved competitive results with considerably shorter required training time in MRBrainsS13 challenge.

Visual inspection of fuel assemblies is necessary to identify potential anomalies in their behaviour associated with their condition and their future usage. One of the possible find- ings are foreign objects caught on the fuel spacer grid which can disrupt the cladding of fuel rods during the operation. The goal of this thesis is to accurately segment the spacer grid from an image, which is a task dual to the foreign object detection, and therefore to automate visual inspection process in this area. We created new datasets covering typical problems appearing on the fuel assembly. To perform the segmentation, we em- ployed neural networks. We increased performance by data augmentation techniques and domain-specific output post-processing. We also measured the algorithm's performance by a newly introduced Line Distance metric, computing the size of the maximum un- certain area between the actual and the predicted transition between grids and rods. In the experiments, we found the best hyperparameters and reached very good results, outperforming our predecessor's algorithm by having three times lower Line Distance metric. 1

Many different types of features can appear on the surface of wooden boards, lineals or parts. Some of these features should not appear on the surfaces of wood products. These features then become undesirable or removable defects for those products. To manufacture these products boards are cutup in such a way that these undesirable defects will not appear in the final product. Studies have shown that manual cutup of boards does not produce the highest possible yield of final product from rough lumber. Because of this fact a good deal of research work has been done to develop automatic defect detection systems. Color images contain a lot of valuable information which can be used to locate and identify features in wood. This is evidenced by the fact that the human color vision system can accurately locate and identify these features. A very important part of any automatic defect detection system based wholly or impart on color imagery is the location of areas that might contain a wood feature, a feature that depending on the product being manufactured may or may not be a defect. This location process is called image segmentation. While a number of automatic defect detection systems have been proposed that employ color imagery, none of these systems use color imagery to do the segmentation. Rather these systems typically average the red, green, and blue color channels together to form a black and white image. The segmentation operation is then performed on the black and white image. The basic hypothesis of this research is that the use of full color imagery to locate defects will yield better segmentation results than can be obtained when only black and white imagery is used. To approach the color wood image segmentation problem, two conventional clustering procedures were selected for examination. Experiments that were performed clearly showed that these procedures, ones that are similar in flavor to other unsupervised clustering methods, are unsuitable for wood color image segmentation. Based on the experience that was gained in examining the unsupervised clustering procedures, a model based approach is developed. This approach is based on the assumption that the distribution of colors in clear wood is Gaussian. Since boards that are used by the forest products secondary manufacturing industry are all such that most of their surface area is clear wood, the idea is to use the most frequently occurring colors, i.e., the ones that must represent the most likely colors of clear wood, to estimate the mean and covariance of the Normal density function specifying the possible colors of clear wood. Deviations from this model in the observed histogram are used to identify colors that must be caused by features other than clear wood that appear on the surface of the board. / Master of Science

Over the past thirty years, X-Ray Computed Tomography (CT) has been widely used in security checking due to its high resolution and fully 3-d construction. Designing object segmentation and classification algorithms based on reconstructed CT intensity data will help accurately locate and classify the potential hazardous articles in luggage. Proposal-based deep networks have been successful recently in segmentation and recognition tasks. However, they require large amount of labeled training images, which are hard to obtain in CT research. This thesis develops a non-proposal 3-d instance segmentation and classification structure based on smoothed fully convolutional networks (FCNs), graph-based spatial clustering and ensembling kernel SVMs using volumetric texture features, which can be trained on limited and highly unbalanced CT intensity data. Our structure will not only significantly accelerate the training convergence in FCN, but also efficiently detect and remove the outlier voxels in training data and guarantee the high and stable material classification performance. We demonstrate the performance of our approach on experimental volumetric images of containers obtained using a medical CT scanner.

No description available.

No description available.

Video segmentation, the process of selecting an object out of a video sequence, is a fundamentally important process for video editing and special effects. However, it remains an unsolved problem due to many difficulties such as large or rapid motions, motion blur, lighting and shadow changes, complex textures, similar colors in the foreground and background, and many others. While the human vision system relies on multiple visual cues and higher-order understanding of the objects involved in order to perceive the segmentation, current algorithms usually depend on a small amount of information to assist a user in selecting a desired object. This causes current methods to often fail for common cases. Because of this, industry still largely relies on humans to trace the object in each frame, a tedious and expensive process. This dissertation investigates methods of segmenting video by propagating the segmentation from frame to frame using multiple cues to maximize the amount of information gained from each user interaction. New and existing methods are incorporated in propagating as much information as possible to a new frame, leveraging multiple cues such as object colors or mixes of colors, color relationships, temporal and spatial coherence, motion, shape, and identifiable points. The cues are weighted and applied on a local basis depending on the reliability of the cue in each region of the image. The reliability of the cues is learned from any corrections the user makes. In this framework, every action of the user is examined and leveraged in an attempt to provide as much information as possible to guarantee a correct segmentation. Propagating segmentation information from frame to frame using multiple cues and learning from the user interaction allows users to more quickly and accurately extract objects from video while exerting less effort.

Geo-specific road database development is important to a driving simulation system and a very labor intensive process. Road databases for driving simulation need high resolution and accuracy. Even though commercial software is available on the market, a lot of manual work still has to be done when the road crosssectional profile is not uniform. This research deals with geo-specific road databases development, especially for roads with non-uniform cross sections. In this research, the United States Geographical Survey (USGS) road information is used with aerial photos to accurately extract road boundaries, using image segmentation and data compression techniques. Image segmentation plays an important role in extracting road boundary information. There are numerous methods developed for image segmentation. Six methods have been tried for the purpose of road image segmentation. The major problems with road segmentation are due to the large variety of road appearances and the many linear features in roads. A method that does not require a database of sample images is desired. Furthermore, this method should be able to handle the complexity of road appearances. The proposed method for road segmentation is based on the mean-shift clustering algorithm and it yields a high accuracy. In the phase of building road databases and visual databases based on road segmentation results, the Linde-Buzo-Gray (LBG) vector quantization algorithm is used to identify repeatable cross section profiles. In the phase of texture mapping, five major uniform textures are considered - pavement, white marker, yellow marker, concrete and grass. They are automatically mapped to polygons. In the chapter of results, snapshots of road/visual database are presented.

Yes / A confocal microscope provides a sequence of images of the various corneal layers and structures at different depths from which medical clinicians can extract clinical information on the state of health of the patient’s cornea. Preprocessing the confocal corneal images to make them suitable for analysis is very challenging due the nature of these images and the amount of the noise present in them. This paper presents an efficient preprocessing approach for confocal corneal images consisting of three main steps including enhancement, binarisation and refinement. Improved visualisation, cell counts and measurements of cell properties have been achieved through this system and an interactive graphical user interface has been developed.