<p>  </p> <p>The inspection of highway bridge structures in the United States is a task critical to the national transportation system. Inspection images contain abundant visual information that can be exploited to streamline bridge assessment and management tasks. However, historical inspection images often go unused in subsequent assessments as they are disorganized and unlabeled. Further, due to the lack of GPS metadata and visual ambiguity, it is often difficult for other inspectors to identify the location on the bridge where past images were taken. While many approaches are being considered toward fully- or semi-automated methods for bridge inspection, there are research opportunities to develop practical tools for inspectors to make use of those images already in a database. In this study, a deep learning-based image similarity technique is combined with image geolocation data to localize and retrieve historical inspection images based on a current query image. A Siamese convolutional neural network (SCNN) is trained and validated on a gathered dataset of over 1,000 real world bridge deck images collected by the Indiana Department of Transportation. A composite similarity (CS) metric is created for effective image ranking and the overall method is validated on a subset of eight bridge’s images. The results show promise for implementation into existing databases and for other similar structural inspections, showing up to an 11-fold improvement in successful image retrieval when compared to random image selection.</p>

There are billions of photos on the internet and as the size of these digital repositories grows, finding target picture becomes more and more difficult. To increase the informational quality of photo albums we propose a new method that selects representative pictures from a group of photographs using computer vision algorithms. The aim of this study is to analyze the issues about image features, image similarity, object clustering and examine the specific characteristics of photographs. Tests show that there is no universal image descriptor that can easily simulate the process of clustering performed by human vision. The thesis proposes a hybrid algorithm that combines the advantages of selected features together using a specialized multiple-step clustering algorithm. The key idea of the process is that the frequently photographed objects are more likely to be representative. Thus, with a random selection from the largest photo clusters certain representative photos are obtained. This selection is further enhanced on the basis of optimization, where photos with better photographic properties are being preferred.

Media consumption has shown a tremendous increase in recent years, and with this increase, new audience expectations are put on the features offered by media-streaming services. One of these expectations is the ability to skip redundant content, which most probably is not of interest to the user. In this work, intro sequences which have sufficient length and a high degree of image similarity across all episodes of a show is targeted for detection. A statistical prediction model for classifying video intros based on these features was proposed. The model tries to identify frame similarities across videos from the same show and then filter out incorrect matches. The performance evaluation of the prediction model shows that the proposed solution for unguided predictions had an accuracy of 90.1%, and precision and recall rate of 93.8% and 95.8% respectively.The mean margin of error for a predicted start and end was 1.4 and 2.0 seconds. The performance was even better if the model had prior knowledge of one or more intro sequences from the same TV series confirmed by a human. However, due to dataset limitations the result is inconclusive. The prediction model was integrated into an automated system for processing internet videos available on SVT Play, and included administrative capabilities for correcting invalid predictions. / Under de senaste åren så har konsumtionen av TV-serier ökat markant och med det tillkommer nya förväntningar på den funktionalitet som erbjuds av webb-TVtjänster. En av dessa förväntningar är förmågan att kunna hoppa över redundant innehåll, vilket troligen inte är av intresse för användaren. I detta arbete så ligger fokus på att detektera video intron som bedöms som tillräckligt långa och har en hög grad av bildlighet över flera episoder från samma TV-program. En statistisk modell för att klassificera intron baserat på dessa egenskaper föreslogs. Modellen jämför bilder från samma TV-program för att försöka identifiera matchande sekvenser och filtrera bort inkorrekta matchningar. Den framtagna modellen hade en träffsäkerhet på 90.1%, precision på 93.8% och en återkallelseförmåga på 95.8%. Medelfelmarginalen uppgick till 1.4 sekunder för start och 2.0 sekunder för slut av ett intro. Modellen presterade bättre om den hade tillgång till en eller fler liknande introsekvenser från relaterade videor från sammaTV-program bekräftat av en människa. Eftersom datasetet som användes för testning hade vissa brister så ska resultatet endast ses som vägledande. Modellen integrerades i ett system som automatiskt processar internet videos frånSVT-Play. Ett tillhörande administrativt verktyg skapades även för att kunna rätta felaktiga gissningar.

Conselho Nacional de Desenvolvimento Científico e Tecnológico / Hoje em dia, a quantidade de usuários que utilizam sites de comércio eletrônico para realizar compras está aumentando muito, principalmente devido à facilidade e rapidez. Muitos sites de comércio eletrônico, diferentemente das lojas físicas, disponibilizam aos seus usuários uma grande variedade de produtos e serviços, e os usuários podem ter muita dificuldade em encontrar produtos de sua preferência. Normalmente, a preferência por um produto pode ser influenciada pela aparência visual da imagem do produto. Neste contexto, os Sistemas de Recomendação de produtos que estão associados a Imagens (SRI) tornaram-se indispensáveis para ajudar os usuários a encontrar produtos que podem ser, eventualmente, agradáveis ou úteis para eles. Geralmente, os SRI usam o comportamento passado dos usuários (cliques, compras, críticas, avaliações, etc.) e/ou atributos de produtos para definirem as preferências dos usuários. Um dos principais desafios enfrentados em SRI é a necessidade de o usuário fornecer algumas informações sobre suas preferências sobre os produtos, a fim de obter novas recomendações do sistema. Infelizmente, os usuários nem sempre estão dispostos a fornecer tais informações de forma explícita. Assim, a fim de lidar com esse desafio, os métodos para obtenção de informações de forma implícita do usuário são desejáveis. Neste trabalho, propõe-se investigar em que medida informações sobre atenção visual do usuário podem ajudar a melhorar a predição de avaliação e consequentemente produzir SRI mais precisos. É também objetivo deste trabalho o desenvolvimento de dois novos métodos, um método baseado em Filtragem Colaborativa (FC) que combina avaliações e dados de atenção visual para representar o comportamento passado dos usuários, e outro método baseado no conteúdo dos itens, que combina atributos textuais, características visuais e dados de atenção visual para compor o perfil dos itens. Os métodos propostos foram avaliados em uma base de imagens de pinturas e uma base de imagens de roupas. Os resultados experimentais mostram que os métodos propostos neste trabalho possuem ganhos significativos em predição de avaliação e precisão na recomendação quando comparados ao estado-da-arte. Vale ressaltar que as técnicas propostas são flexíveis, podendo ser utilizadas em outros cenários que exploram a atenção visual dos itens recomendados. / Nowadays, the amount of users using e-commerce sites for shopping is greatly increasing, mainly due to the easiness and rapidity of this way of consumption. Many e-commerce sites, differently from physical stores, can offer their users a wide range of products and services, and the users can find it very difficult to find products of your preference. Typically, your preference for a product can be influenced by the visual appearance of the product image. In this context, Image Recommendation Systems (IRS) have become indispensable to help users to find products that may possibly pleasant or be useful to them. Generally, IRS use past behavior of users (clicks, purchases, reviews, ratings, etc.) and/or attributes of the products to define the preferences of users. One of the main challenges faced by IRS is the need of the user to provide some information about his / her preferences on products in order to get further recommendations from the system. Unfortunately, users are not always willing to provide such information explicitly. So, in order to cope with this challenge, methods for obtaining user’s implicit feedback are desirable. In this work, the author propose an investigation to discover to which extent information concerning user visual attention can help improve the rating prediction hence produce more accurate IRS. This work proposes to develop two new methods, a method based on Collaborative Filtering (CF) which combines ratings and data visual attention to represent the past behavior of users, and another method based on the content of the items, which combines textual attributes, visual features and visual attention data to compose the profile of the items. The proposed methods were evaluated in a painting dataset and a clothing dataset. The experimental results show significant improvements in rating prediction and precision in recommendation when compared to the state-of-the-art. It is worth mentioning that the proposed techniques are flexible and can be applied in other scenarios that exploits the visual attention of the recommended items. / Tese (Doutorado)

Thesis in theoretical part deals with the procedures used in image databases searching. There are discussed two basic possible approaches - text based searching and content based searching. In next section there are described methods for image similarity detection. Practical part deals with detailed description and implementation of three selected image features used for image searching. In third part there are presented testing procedure for implemented algorithms and test results. In conclusion implementation of Rapidminer operator are described. This operator uses all implemented algorithms and allows image similarity matching, searching for most similar images in database, and copy these images to output folder.

This work focuses on a possibility of using current computer vision alghoritms and methods for automatic similarity matching of so called graffiti tags. Those are such graffiti, that are used as a fast and simple signature of their authors. The process of development and implementation of CBIR system, which is created for this task, is described. For the purposes of finding images similarity, local features are used, most notably self-similarity features.