Campos Aleatórios Condicionais são modelos probabilísticos discriminativos que tem sido utilizados com sucesso em diversas áreas como processamento de linguagem natural, reconhecimento de fala e bioinformática. Entretanto, implementar algoritmos eficientes para esse tipo de modelo não é uma tarefa fácil. Nesse trabalho apresentamos um arcabouço que ajuda no desenvolvimento e experimentação de Campos Aleatórios Condicionais Semi Markovianos (semi-CRFs). Desenvolvemos algoritmos eficientes que foram implementados em C++ propondo uma interface de programação flexível e intuitiva que habilita o usuário a definir, treinar e avaliar modelos. Nossa implementação foi construída como uma extensão do arcabouço ToPS que, inclusive, pode utilizar qualquer modelo já definido no ToPS como uma função de característica especializada. Por fim utilizamos nossa implementação de semi-CRF para construir um preditor de promotores que apresentou performance superior aos preditores existentes. / Conditional Random Fields are discriminative probabilistic models that have been successfully used in several areas like natural language processing, speech recognition and bioinformatics. However, implementing efficient algorithms for this kind of model is not an easy task. In this thesis we show a framework that helps the development and experimentation of Semi-Markov Conditional Random Fields (semi-CRFs). It has an efficient implementation in C++ and an intuitive API that allow users to define, train and evaluate models. It was built as an extension of ToPS framework and can use ToPS probabilistic models as specialized feature functions. We also use our implementation of semi-CRFs to build a high performance promoter predictor.

" A novel Markov Random Field (MRF) based method for the mosaicing of 3D ultrasound volumes is presented in this dissertation. The motivation for this work is the production of training volumes for an affordable ultrasound simulator, which offers a low-cost/portable training solution for new users of diagnostic ultrasound, by providing the scanning experience essential for developing the necessary psycho-motor skills. It also has the potential for introducing ultrasound instruction into medical education curriculums. The interest in ultrasound training stems in part from the widespread adoption of point-of-care scanners, i.e. low cost portable ultrasound scanning systems in the medical community. This work develops a novel approach for producing 3D composite image volumes and validates the approach using clinically acquired fetal images from the obstetrics department at the University of Massachusetts Medical School (UMMS). Results using the Visible Human Female dataset as well as an abdominal trauma phantom are also presented. The process is broken down into five distinct steps, which include individual 3D volume acquisition, rigid registration, calculation of a mosaicing function, group-wise non-rigid registration, and finally blending. Each of these steps, common in medical image processing, has been investigated in the context of ultrasound mosaicing and has resulted in improved algorithms. Rigid and non-rigid registration methods are analyzed in a probabilistic framework and their sensitivity to ultrasound shadowing artifacts is studied. The group-wise non-rigid registration problem is initially formulated as a maximum likelihood estimation, where the joint probability density function is comprised of the partially overlapping ultrasound image volumes. This expression is simplified using a block-matching methodology and the resulting discrete registration energy is shown to be equivalent to a Markov Random Field. Graph based methods common in computer vision are then used for optimization, resulting in a set of transformations that bring the overlapping volumes into alignment. This optimization is parallelized using a fusion approach, where the registration problem is divided into 8 independent sub-problems whose solutions are fused together at the end of each iteration. This method provided a speedup factor of 3.91 over the single threaded approach with no noticeable reduction in accuracy during our simulations. Furthermore, the registration problem is simplified by introducing a mosaicing function, which partitions the composite volume into regions filled with data from unique partially overlapping source volumes. This mosaicing functions attempts to minimize intensity and gradient differences between adjacent sources in the composite volume. Experimental results to demonstrate the performance of the group-wise registration algorithm are also presented. This algorithm is initially tested on deformed abdominal image volumes generated using a finite element model of the Visible Human Female to show the accuracy of its calculated displacement fields. In addition, the algorithm is evaluated using real ultrasound data from an abdominal phantom. Finally, composite obstetrics image volumes are constructed using clinical scans of pregnant subjects, where fetal movement makes registration/mosaicing especially difficult. Our solution to blending, which is the final step of the mosaicing process, is also discussed. The trainee will have a better experience if the volume boundaries are visually seamless, and this usually requires some blending prior to stitching. Also, regions of the volume where no data was collected during scanning should have an ultrasound-like appearance before being displayed in the simulator. This ensures the trainee's visual experience isn't degraded by unrealistic images. A discrete Poisson approach has been adapted to accomplish these tasks. Following this, we will describe how a 4D fetal heart image volume can be constructed from swept 2D ultrasound. A 4D probe, such as the Philips X6-1 xMATRIX Array, would make this task simpler as it can acquire 3D ultrasound volumes of the fetal heart in real-time; However, probes such as these aren't widespread yet. Once the theory has been introduced, we will describe the clinical component of this dissertation. For the purpose of acquiring actual clinical ultrasound data, from which training datasets were produced, 11 pregnant subjects were scanned by experienced sonographers at the UMMS following an approved IRB protocol. First, we will discuss the software/hardware configuration that was used to conduct these scans, which included some custom mechanical design. With the data collected using this arrangement we generated seamless 3D fetal mosaics, that is, the training datasets, loaded them into our ultrasound training simulator, and then subsequently had them evaluated by the sonographers at the UMMS for accuracy. These mosaics were constructed from the raw scan data using the techniques previously introduced. Specific training objectives were established based on the input from our collaborators in the obstetrics sonography group. Important fetal measurements are reviewed, which form the basis for training in obstetrics ultrasound. Finally clinical images demonstrating the sonographer making fetal measurements in practice, which were acquired directly by the Philips iU22 ultrasound machine from one of our 11 subjects, are compared with screenshots of corresponding images produced by our simulator. "

Wu, Yiu Kei. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2008. / Includes bibliographical references (p. 69-75). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Chinese NER Problem --- p.1 / Chapter 1.2 --- Contribution of Our Proposed Framework --- p.3 / Chapter 2 --- Related Work --- p.6 / Chapter 2.1 --- Hidden Markov Models --- p.7 / Chapter 2.2 --- Maximum Entropy Models --- p.8 / Chapter 2.3 --- Conditional Random Fields --- p.10 / Chapter 3 --- Our Proposed Model --- p.14 / Chapter 3.1 --- Background --- p.14 / Chapter 3.1.1 --- Problem Formulation --- p.14 / Chapter 3.1.2 --- Conditional Random Fields --- p.16 / Chapter 3.1.3 --- Semi-Markov Conditional Random Fields --- p.26 / Chapter 3.2 --- The Formulation of Our Proposed Model --- p.28 / Chapter 3.2.1 --- The Main Principle --- p.28 / Chapter 3.2.2 --- The Detailed Formulation --- p.36 / Chapter 3.2.3 --- Adapting Features from Original CRF to CRFDP --- p.51 / Chapter 4 --- Experiments --- p.54 / Chapter 4.1 --- Datasets --- p.55 / Chapter 4.2 --- Features --- p.57 / Chapter 4.3 --- Evaluation Metrics --- p.61 / Chapter 4.4 --- Results and Discussion --- p.63 / Chapter 5 --- Conclusions and Future Work --- p.67 / Bibliography --- p.69 / A --- p.76 / B --- p.78 / C --- p.88

We seek to both detect and segment objects in images. To exploit both local image data as well as contextual information, we introduce Boosted Random Fields (BRFs), which uses Boosting to learn the graph structure and local evidence of a conditional random field (CRF). The graph structure is learned by assembling graph fragments in an additive model. The connections between individual pixels are not very informative, but by using dense graphs, we can pool information from large regions of the image; dense models also support efficient inference. We show how contextual information from other objects can improve detection performance, both in terms of accuracy and speed, by using a computational cascade. We apply our system to detect stuff and things in office and street scenes.

There is significant interest in the synthesis of discrete-state random fields, particularly those possessing structure over a wide range of scales. However, given a model on some finest, pixellated scale, it is computationally very difficult to synthesize both large and small-scale structures, motivating research into hierarchical methods. This thesis proposes a frozen-state approach to hierarchical modelling, in which simulated annealing is performed on each scale, constrained by the state estimates at the parent scale. The approach leads significant advantages in both modelling flexibility and computational complexity. In particular, a complex structure can be realized with very simple, local, scale-dependent models, and by constraining the domain to be annealed at finer scales to only the uncertain portions of coarser scales, the approach leads to huge improvements in computational complexity. Results are shown for synthesis problems in porous media.

In Chinese language, the smallest meaningful unit is a word which is composed of a sequence of characters. A Chinese sentence is composed of a sequence of words without any separation between them. In the area of information retrieval or data mining, the segmentation of a sequence of Chinese characters should be done before anyone starts to use these segments of characters. The process is called the Chinese word segmentation. The researches of Chinese word segmentation have been developed for many years. Although some recent researches have achieved very high performance, the recall of those words that are not in the dictionary only achieves sixty or seventy percent. An approach described in this paper makes use of the linear-chain conditional random fields (CRFs) to have a more accurate Chinese word segmentation. The discriminatively trained model that uses two of our proposed feature templates for deciding the boundaries between characters is used in our study. We also propose three other methods, which are the duplicate word repartition, the date representation repartition, and the segment refinement, to enhance the accuracy of the processed segments. In the experiments, we use several different approaches for testing and compare the results with those proposed by Li et al. and Lau and King based on three different Chinese word corpora. The results prove that the improved feature template which makes use of the information of prefix and postfix could increase both the recall and the precision. For example, the F-measure reaches 0.964 in the MSR dataset. By detecting repeat characters, the duplicated characters could also be better repartitioned without using extra resources. In the representation of date, the wrongly segmented date could be better repartitioned by using the proposed method which deals with numbers, date, and measure words. If a word is segmented differently from that of the corresponding standard segmentation corpus, a proper segment could be produced by repartitioning the assembled segment which is composed of the current segment and the adjacent segment. In the area of using the conditional random fields for Chinese word segmentation, we have proposed a feature template for better result and three methods which focus on other specific segmentation problems.

Thesis (Ph. D.)--Washington State University, December 2008. / Title from PDF title page (viewed on Sept. 23, 2008) "School of Electrical Engineering and Computer Science." Includes bibliographical references (p. 83-90).

Thesis (M. Phil.)--University of Hong Kong, 2010. / Includes bibliographical references (leaves 54-59). Also available in print.

Great progress has been made in face sketch synthesis in recent years. State-of-the-art methods commonly apply a Markov Random Fields (MRF) model to select local sketch patches from a set of training data. Such methods, however, have two major drawbacks. Firstly, the MRF model used cannot synthesize new sketch patches. Secondly, the optimization problem in solving the MRF is NP-hard. In this thesis, a novel Markov Weight Fields (MWF) model is proposed. By applying linear combination of candidate patches, MWF is capable of synthesizing new sketch patches. The MWF model can be formulated into a convex quadratic programming (QP) problem to which the optimal solution is guaranteed. Based on the Markov property of MWF model, a cascade decomposition method (CDM) is further proposed for solving such a large scale QP problem efficiently. Experiments show that the proposed CDM is very efficient, and only takes about 2:4 seconds. To deal with illumination changes of input photos, five special shading patches are included as candidate patches in addition to the patches selected from the training data. These patches help keeping structure of the face under different illumination conditions as well as synthesize shadows similar to the input photos. Extensive experiments on the CUHK face sketch database, AR database and Chinese celebrity photos show that the proposed model outperforms the common MRF model used in other state-of-the-art methods and is robust to illumination changes. / published_or_final_version / Computer Science / Master / Master of Philosophy

No description available.