Sistemas para análise de imagens partem da premissa de que o conjunto de dados sob investigação está corretamente representado por características. Entretanto, definir quais características representam apropriadamente um conjunto de dados é uma tarefa desafiadora e exaustiva. Grande parte das técnicas de descrição existentes na literatura, especialmente quando os dados têm alta dimensionalidade, são baseadas puramente em medidas estatísticas ou abordagens baseadas em inteligência artificial, e normalmente são caixas-pretas para os usuários. A abordagem proposta nesta dissertação busca abrir esta caixa-preta por meio de representações visuais criadas pela técnica Multidimensional Classical Scaling, permitindo que usuários capturem interativamente a essência sobre a representatividade das características computadas de diferentes descritores. A abordagem é avaliada sobre seis conjuntos de imagens que contém texturas, imagens médicas e cenas naturais. Os experimentos mostram que, conforme a combinação de um conjunto de características melhora a qualidade da representação visual, a acurácia de classificação também melhora. A qualidade das representações é medida pelo índice da silhueta, superando problemas relacionados com a subjetividade de conclusões baseadas puramente em análise visual. Além disso, a capacidade de exploração visual do conjunto sob análise permite que usuários investiguem um dos maiores desafios em classificação de dados: a presença de variação intra-classe. Os resultados sugerem fortemente que esta abordagem pode ser empregada com sucesso como um guia para auxiliar especialistas a explorar, refinar e definir as características que representam apropriadamente um conjunto de imagens / Image analysis systems rely on the fact that the dataset under investigation is correctly represented by features. However, defining a set of features that properly represents a dataset is still a challenging and, in most cases, an exhausting task. Most of the available techniques, especially when a large number of features is considered, are based on purely quantitative statistical measures or approaches based on artificial intelligence, and normally are black-boxes to the user. The approach proposed here seeks to open this black-box by means of visual representations via Multidimensional Classical Scaling projection technique, enabling users to get insight about the meaning and representativeness of the features computed from different feature extraction algorithms and sets of parameters. This approach is evaluated over six image datasets that contains textures, medical images and outdoor scenes. The results show that, as the combination of sets of features and changes in parameters improves the quality of the visual representation, the accuracy of the classification for the computed features also improves. In order to reduce this subjectiveness, a measure called silhouette index, which was originally proposed to evaluate results of clustering algorithms, is employed. Moreover, the visual exploration of datasets under analysis enable users to investigate one of the greatest challenges in data classification: the presence of intra-class variation. The results strongly suggest that our approach can be successfully employed as a guidance to defining and understanding a set of features that properly represents an image dataset

Sistemas para análise de imagens partem da premissa de que o conjunto de dados sob investigação está corretamente representado por características. Entretanto, definir quais características representam apropriadamente um conjunto de dados é uma tarefa desafiadora e exaustiva. Grande parte das técnicas de descrição existentes na literatura, especialmente quando os dados têm alta dimensionalidade, são baseadas puramente em medidas estatísticas ou abordagens baseadas em inteligência artificial, e normalmente são caixas-pretas para os usuários. A abordagem proposta nesta dissertação busca abrir esta caixa-preta por meio de representações visuais criadas pela técnica Multidimensional Classical Scaling, permitindo que usuários capturem interativamente a essência sobre a representatividade das características computadas de diferentes descritores. A abordagem é avaliada sobre seis conjuntos de imagens que contém texturas, imagens médicas e cenas naturais. Os experimentos mostram que, conforme a combinação de um conjunto de características melhora a qualidade da representação visual, a acurácia de classificação também melhora. A qualidade das representações é medida pelo índice da silhueta, superando problemas relacionados com a subjetividade de conclusões baseadas puramente em análise visual. Além disso, a capacidade de exploração visual do conjunto sob análise permite que usuários investiguem um dos maiores desafios em classificação de dados: a presença de variação intra-classe. Os resultados sugerem fortemente que esta abordagem pode ser empregada com sucesso como um guia para auxiliar especialistas a explorar, refinar e definir as características que representam apropriadamente um conjunto de imagens / Image analysis systems rely on the fact that the dataset under investigation is correctly represented by features. However, defining a set of features that properly represents a dataset is still a challenging and, in most cases, an exhausting task. Most of the available techniques, especially when a large number of features is considered, are based on purely quantitative statistical measures or approaches based on artificial intelligence, and normally are black-boxes to the user. The approach proposed here seeks to open this black-box by means of visual representations via Multidimensional Classical Scaling projection technique, enabling users to get insight about the meaning and representativeness of the features computed from different feature extraction algorithms and sets of parameters. This approach is evaluated over six image datasets that contains textures, medical images and outdoor scenes. The results show that, as the combination of sets of features and changes in parameters improves the quality of the visual representation, the accuracy of the classification for the computed features also improves. In order to reduce this subjectiveness, a measure called silhouette index, which was originally proposed to evaluate results of clustering algorithms, is employed. Moreover, the visual exploration of datasets under analysis enable users to investigate one of the greatest challenges in data classification: the presence of intra-class variation. The results strongly suggest that our approach can be successfully employed as a guidance to defining and understanding a set of features that properly represents an image dataset

No description available.

Since last two decades researches have been working on developing systems that can assistsdrivers in the best way possible and make driving safe. Computer vision has played a crucialpart in design of these systems. With the introduction of vision techniques variousautonomous and robust real-time traffic automation systems have been designed such asTraffic monitoring, Traffic related parameter estimation and intelligent vehicles. Among theseautomatic detection and recognition of road signs has became an interesting research topic.The system can assist drivers about signs they don’t recognize before passing them.Aim of this research project is to present an Intelligent Road Sign Recognition System basedon state-of-the-art technique, the Support Vector Machine. The project is an extension to thework done at ITS research Platform at Dalarna University [25]. Focus of this research work ison the recognition of road signs under analysis. When classifying an image its location, sizeand orientation in the image plane are its irrelevant features and one way to get rid of thisambiguity is to extract those features which are invariant under the above mentionedtransformation. These invariant features are then used in Support Vector Machine forclassification. Support Vector Machine is a supervised learning machine that solves problemin higher dimension with the help of Kernel functions and is best know for classificationproblems.

In recent years, many researches focused on the supervised machine learning classification methods using Lidar and remotely sensed image to provide buildings, trees, roads, and grass categories for urban ground feature classification. First, this research performed urban ground feature classification based on true color aerial imagey and Lidar Intensity. Second, Lidar derived normalized DSM (nDSM) was added to the classification. Finally, the concept of height level rules was applied. This research utilized two-level height rule-based classification exteneded from three-level height rule-based classification (Huang, 2007). It is obvious to observ the overlap for the roads and houses, and grass and trees in the feature space plot where result in the classification confusion. These confusions can be resolved by fusion the height information. After comparing classification accuracy, the two-level height is better than three-level height classification scheme. This research proposed hybrid classification method based on Maximum likelihood classification (MLC) and two-level height rules. This method reveals the role of height information in urban ground feature classification. The height level rules were also applied to other supervised classification method such as Back-Propagation Network (BPN) and Support Vector Machine (SVM). The classification results show that the accuracy of hybrid method is better than the orgional classification method. However, the time required to look for the classification parameters for BPN and SVM is greater than MLC but only can derived considerable results. Therefore, the hybrid classification method based on MLC is better than other two methods.

Behavior recognition provides an interesting perspective for understandingthe different modes of a system and the influence of eachmode under varying conditions. In most of the systems, prior knowledgeof different expected behavior is available. Whereas, in an automotivedomain, a fleet of vehicle with many external factors influencingeach vehicle and an asynchronous performance of each vehicleon road, creates the complexity on analyzing and predicting the exacttime segments of vehicles in a fleet exhibiting similar behavior. Thisthesis focuses on recognizing time segments of vehicles that exhibitsimilar behavior based on supervised and unsupervised approaches.In supervised approach, classifiers are trained to predict two distinctiveoperations(highway and in-city). In unsupervised approach, featurespace is explored for identification of consistent features and existenceof other operations. An unsupervised approach to recognizepeer cluster groups is combined with supervised classification resultsto achieve lower computational complexity.

The goal of this study is to develop a smoke detecting algorithm using digital image processing techniques on multi-spectral (visible & infrared) video. By utilizing principal component analysis (PCA) followed by spatial filtering of principal component images the location of smoke can be accurately identified over a period of exposure time with a given frame capture rate. This result can be further analyzed with consideration of wind factor and fire detection range to determine if a fire is present within a scene. Infrared spectral data is shown to contribute little information concerning the smoke signature. Moreover, finalized processing techniques are focused on the blue spectral band as it is furthest away from the infrared spectral bands and because it experimentally yields the largest footprint in the processed principal component images in comparison to other spectral bands. A frame rate of .5 images/sec (1 image every 2 seconds) is determined to be the maximum such that temporal variance of smoke can be captured. The study also shows eigenvectors corresponding to the principal components that best represent smoke and are valuable indications of smoke temporal signature. Raw video data is taken through rigorous pre-processing schemes to align frames from respective spectral band both spatially and temporally. A multi-paradigm numerical computing program, MATLAB, is used to match the field of view across five spectral bands: Red, Green, Blue, Long-Wave Infrared, and Mid-Wave Infrared. Extracted frames are aligned temporally from key frames throughout the data capture. This alignment allows for more accurate digital processing for smoke signature. v Clustering analysis on RGB and HSV value systems reveal that color alone is not helpful to segment smoke. The feature values of trees and other false positives are shown to be too closely related to features of smoke for in solely one instance in time. A temporal principal component transform on the blue spectral band eliminates static false positives and emphasizes the temporal variance of moving smoke in images with higher order. A threshold adjustment is applied to a blurred blue principal component of non-unity principal component order and smoke results can be finalized using median filtering. These same processing techniques are applied to difference images as a more simple and traditional technique for identifying temporal variance and results are compared.

This thesis first describes theory of range of methods of textural and shape analysis. In several published articles some of the mentioned methods are used for automatic detection of lesion in spine in CT images. Some of these articles are shortly presented (in this thesis). Next part of the thesis includes description of various classifiers which are used for classification of feature vectors. Practical part of the thesis is a design and implementation of image data segmentation solution (metastatic lesions in vertebrae) with use of classification of feature vectors formed by texture and shape symptoms. The thesis also deals with the selection of significant features for segmentation. Segmentation algorithm is tested on medical data.

Semantic segmentation of a scene aims to give meaning to the scene by dividing it into meaningful — semantic — parts. Understanding the scene is of great interest for all kinds of autonomous systems, but manual annotation is simply too time consuming, which is why there is a need for an alternative approach. This thesis investigates the possibility of automatically segmenting 3D-models of urban scenes, such as buildings, into a predetermined set of labels. The approach was to first acquire ground truth data by manually annotating five 3D-models of different urban scenes. The next step was to extract features from the 3D-models and evaluate which ones constitutes a suitable feature space. Finally, three supervised learners were implemented and evaluated: k-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Random Classification Forest (RCF). The classifications were done point-wise, classifying each 3D-point in the dense point cloud belonging to the model being classified. The result showed that the best suitable feature space is not necessarily the one containing all features. The KNN classifier got the highest average accuracy overall models — classifying 42.5% of the 3D points correct. The RCF classifier managed to classify 66.7% points correct in one of the models, but had worse performance for the rest of the models and thus resulting in a lower average accuracy compared to KNN. In general, KNN, SVM, and RCF seemed to have different benefits and drawbacks. KNN is simple and intuitive but by far the slowest classifier when dealing with a large set of training data. SVM and RCF are both fast but difficult to tune as there are more parameters to adjust. Whether the reason for obtaining the relatively low highest accuracy was due to the lack of ground truth training data, unbalanced validation models, or the capacity of the learners, was never investigated due to a limited time span. However, this ought to be investigated in future studies.

Na maioria dos processos de análise de imagens há a necessidade de um pré-processamento, no qual são extraídos e calculados vetores de características que representem as imagens são utilizados no cálculo de similaridade. Uma dificuldade nessas tarefas é o grande número de características que definem um espaço de alta dimensionalidade, afetando fortemente o desempenho das tarefas que seguem, que podem envolver uma análise visual, um agrupamento ou uma classificação de dados, por exemplo. Lidar com esse problema normalmente exige técnicas de redução de dimensionalidade ou seleção de características. O presente trabalho dá sequência a trabalhos que utilizam técnicas de visualização como suporte para avaliar espaços de características gerados a partir de coleções de imagens. Nele, objetiva-se aprimorar um método baseado na análise visual de conjuntos de imagens empregando a árvore de similaridade Neighbor-Joining que apoia o usuário a selecionar um subespaço de características que mantenha ou melhore os resultados das visualizações do conjunto de imagens. A partir da metodologia proposta, a avaliação e a seleção de características representativas é realizada usando a visualização NJ. A maior parte dos experimentos responde positivamente para diferentes conjuntos de imagens representados por vários extratores, obtendo-se processos de seleção personalizados mais precisos e eficazes, em termos de agrupamento, do que abordagens automáticas reportadas na literatura / In the majority of the image analysis processes there is need for a pre-processing step, in which feature vectors representative of the images are extracted and similarity methods are calculates. A difficult step in the process is to choose amongst the large number of features available, that will define a feature space of high dimensionality, impacting the cost of the subsequent processing tasks, such as visual analysis, clustering and classification. This problem is usually handled by dimension reduction of feature selection techniques. This work extends and improves previous work that employs visualization and visual analysis techniques to support evaluation of feature spaces created from image collections. The goal is to improve a previous method of feature selection through visualization to employ similarity trees via the Neighbor Joining (NJ) algorithm as the basis for the visual layout, as well as to improve the choices of the analyst regarding tools for visual selection of features. The same process can be employed to support evaluation of feature spaces using the NJ visualization. The majorities of experiments results in improvement of spaces generated by various extractors, yielding personalized selection process that are more precisely related to user\'s perspective of the data set and are perform similarly or better than automatic approaches available in the literature. Keywords: information visualization, mining, visual images, visual analysis of the feature space, similarity trees