A contribution to semantic description of images and videos: an application of soft biometrics

Humans have a high ability to extract visual data information acquired by sight. Trought a learning process, which starts at birth and continues throughout life, image interpretation becomes almost instinctively. At a glance, one can easily describe a scene with reasonable precision, naming its main components. Usually, this is done by extracting low-level features such as edges, shapes and textures, and associanting them to high level meanings. In this way, a semantic description of the scene is done. An example of this, is the human capacity to recognize and describe other people physical and behavioral characteristics, or biometrics. Soft-biometrics also represents inherent characteristics of human body and behaviour, but do not allow unique person identification. Computer vision area aims to develop methods capable of performing visual interpretation with performance similar to humans. This thesis aims to propose computer vison methods which allows high level information extraction from images in the form of soft biometrics. This problem is approached in two ways, unsupervised and supervised learning methods. The first seeks to group images via an automatic feature extraction learning , using both convolution techniques, evolutionary computing and clustering. In this approach employed images contains faces and people. Second approach employs convolutional neural networks, which have the ability to operate on raw images, learning both feature extraction and classification processes. Here, images are classified according to gender and clothes, divided into upper and lower parts of human body. First approach, when tested with different image datasets obtained an accuracy of approximately 80% for faces and non-faces and 70% for people and non-person. The second tested using images and videos, obtained an accuracy of about 70% for gender, 80% to the upper clothes and 90% to lower clothes. The results of these case studies, show that proposed methods are promising, allowing the realization of automatic high level information image annotation. This opens possibilities for development of applications in diverse areas such as content-based image and video search and automatica video survaillance, reducing human effort in the task of manual annotation and monitoring.

The role of context in image annotation and recommendation

With the rise of smart phones, lifelogging devices (e.g. Google Glass) and popularity of image sharing websites (e.g. Flickr), users are capturing and sharing every aspect of their life online producing a wealth of visual content. Of these uploaded images, the majority are poorly annotated or exist in complete semantic isolation making the process of building retrieval systems difficult as one must firstly understand the meaning of an image in order to retrieve it. To alleviate this problem, many image sharing websites offer manual annotation tools which allow the user to “tag” their photos, however, these techniques are laborious and as a result have been poorly adopted; Sigurbjörnsson and van Zwol (2008) showed that 64% of images uploaded to Flickr are annotated with < 4 tags. Due to this, an entire body of research has focused on the automatic annotation of images (Hanbury, 2008; Smeulders et al., 2000; Zhang et al., 2012a) where one attempts to bridge the semantic gap between an image’s appearance and meaning e.g. the objects present. Despite two decades of research the semantic gap still largely exists and as a result automatic annotation models often offer unsatisfactory performance for industrial implementation. Further, these techniques can only annotate what they see, thus ignoring the “bigger picture” surrounding an image (e.g. its location, the event, the people present etc). Much work has therefore focused on building photo tag recommendation (PTR) methods which aid the user in the annotation process by suggesting tags related to those already present. These works have mainly focused on computing relationships between tags based on historical images e.g. that NY and timessquare co-exist in many images and are therefore highly correlated. However, tags are inherently noisy, sparse and ill-defined often resulting in poor PTR accuracy e.g. does NY refer to New York or New Year? This thesis proposes the exploitation of an image’s context which, unlike textual evidences, is always present, in order to alleviate this ambiguity in the tag recommendation process. Specifically we exploit the “what, who, where, when and how” of the image capture process in order to complement textual evidences in various photo tag recommendation and retrieval scenarios. In part II, we combine text, content-based (e.g. # of faces present) and contextual (e.g. day-of-the-week taken) signals for tag recommendation purposes, achieving up to a 75% improvement to precision@5 in comparison to a text-only TF-IDF baseline. We then consider external knowledge sources (i.e. Wikipedia & Twitter) as an alternative to (slower moving) Flickr in order to build recommendation models on, showing that similar accuracy could be achieved on these faster moving, yet entirely textual, datasets. In part II, we also highlight the merits of diversifying tag recommendation lists before discussing at length various problems with existing automatic image annotation and photo tag recommendation evaluation collections. In part III, we propose three new image retrieval scenarios, namely “visual event summarisation”, “image popularity prediction” and “lifelog summarisation”. In the first scenario, we attempt to produce a rank of relevant and diverse images for various news events by (i) removing irrelevant images such memes and visual duplicates (ii) before semantically clustering images based on the tweets in which they were originally posted. Using this approach, we were able to achieve over 50% precision for images in the top 5 ranks. In the second retrieval scenario, we show that by combining contextual and content-based features from images, we are able to predict if it will become “popular” (or not) with 74% accuracy, using an SVM classifier. Finally, in chapter 9 we employ blur detection and perceptual-hash clustering in order to remove noisy images from lifelogs, before combining visual and geo-temporal signals in order to capture a user’s “key moments” within their day. We believe that the results of this thesis show an important step towards building effective image retrieval models when there lacks sufficient textual content (i.e. a cold start).

The Use of Latent Semantic Indexing to Mitigate OCR Effects of Related Document Images

Due to both the widespread and multipurpose use of document images and the current availability of a high number of document images repositories, robust information retrieval mechanisms and systems have been increasingly demanded. This paper presents an approach to support the automatic generation of relationships among document images by exploiting Latent Semantic Indexing (LSI) and Optical Character Recognition (OCR). We developed the LinkDI (Linking of Document Images) service, which extracts and indexes document images content, computes its latent semantics, and defines relationships among images as hyperlinks. LinkDI was experimented with document images repositories, and its performance was evaluated by comparing the quality of the relationships created among textual documents as well as among their respective document images. Considering those same document images, we ran further experiments in order to compare the performance of LinkDI when it exploits or not the LSI technique. Experimental results showed that LSI can mitigate the effects of usual OCR misrecognition, which reinforces the feasibility of LinkDI relating OCR output with high degradation.

A synchronous multimedia annotation system for secure collaboratories

In this paper, we describe the Vannotea system - an application designed to enable collaborating groups to discuss and annotate collections of high quality images, video, audio or 3D objects. The system has been designed specifically to capture and share scholarly discourse and annotations about multimedia research data by teams of trusted colleagues within a research or academic environment. As such, it provides: authenticated access to a web browser search interface for discovering and retrieving media objects; a media replay window that can incorporate a variety of embedded plug-ins to render different scientific media formats; an annotation authoring, editing, searching and browsing tool; and session logging and replay capabilities. Annotations are personal remarks, interpretations, questions or references that can be attached to whole files, segments or regions. Vannotea enables annotations to be attached either synchronously (using jabber message passing and audio/video conferencing) or asynchronously and stand-alone. The annotations are stored on an Annotea server, extended for multimedia content. Their access, retrieval and re-use is controlled via Shibboleth identity management and XACML access policies.

An Approach To Publish a Data Warehouse Content as Linked Data

Mestrado em Engenharia Informática - Área de Especialização em Tecnologias do Conhecimento e Decisão

Automated image tagging through tag propagation

Trabalho apresentado no âmbito do Mestrado em Engenharia Informática, como requisito parcial Para obtenção do grau de Mestre em Engenharia Informática

Gesture based interface for image annotation

Dissertação apresentada para obtenção do Grau de Mestre em Engenharia Informática pela Universidade Nova de Lisboa, Faculdade de Ciências e Tecnologia

People and object tracking for video annotation

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

Video metadata extraction in a videoMail system

Currently the world swiftly adapts to visual communication. Online services like YouTube and Vine show that video is no longer the domain of broadcast television only. Video is used for different purposes like entertainment, information, education or communication. The rapid growth of today’s video archives with sparsely available editorial data creates a big problem of its retrieval. The humans see a video like a complex interplay of cognitive concepts. As a result there is a need to build a bridge between numeric values and semantic concepts. This establishes a connection that will facilitate videos’ retrieval by humans. The critical aspect of this bridge is video annotation. The process could be done manually or automatically. Manual annotation is very tedious, subjective and expensive. Therefore automatic annotation is being actively studied. In this thesis we focus on the multimedia content automatic annotation. Namely the use of analysis techniques for information retrieval allowing to automatically extract metadata from video in a videomail system. Furthermore the identification of text, people, actions, spaces, objects, including animals and plants. Hence it will be possible to align multimedia content with the text presented in the email message and the creation of applications for semantic video database indexing and retrieving.

Does Semantic Context Benefit Speech Understanding through "Top-Down" Processes? Evidence from Time-resolved Sparse fMRI.

When speech is degraded, word report is higher for semantically coherent sentences (e.g., her new skirt was made of denim) than for anomalous sentences (e.g., her good slope was done in carrot). Such increased intelligibility is often described as resulting from "top-down" processes, reflecting an assumption that higher-level (semantic) neural processes support lower-level (perceptual) mechanisms. We used time-resolved sparse fMRI to test for top-down neural mechanisms, measuring activity while participants heard coherent and anomalous sentences presented in speech envelope/spectrum noise at varying signal-to-noise ratios (SNR). The timing of BOLD responses to more intelligible speech provides evidence of hierarchical organization, with earlier responses in peri-auditory regions of the posterior superior temporal gyrus than in more distant temporal and frontal regions. Despite Sentence content × SNR interactions in the superior temporal gyrus, prefrontal regions respond after auditory/perceptual regions. Although we cannot rule out top-down effects, this pattern is more compatible with a purely feedforward or bottom-up account, in which the results of lower-level perceptual processing are passed to inferior frontal regions. Behavioral and neural evidence that sentence content influences perception of degraded speech does not necessarily imply "top-down" neural processes.

The UniProt-GO Annotation database in 2011.

The GO annotation dataset provided by the UniProt Consortium (GOA: http://www.ebi.ac.uk/GOA) is a comprehensive set of evidenced-based associations between terms from the Gene Ontology resource and UniProtKB proteins. Currently supplying over 100 million annotations to 11 million proteins in more than 360,000 taxa, this resource has increased 2-fold over the last 2 years and has benefited from a wealth of checks to improve annotation correctness and consistency as well as now supplying a greater information content enabled by GO Consortium annotation format developments. Detailed, manual GO annotations obtained from the curation of peer-reviewed papers are directly contributed by all UniProt curators and supplemented with manual and electronic annotations from 36 model organism and domain-focused scientific resources. The inclusion of high-quality, automatic annotation predictions ensures the UniProt GO annotation dataset supplies functional information to a wide range of proteins, including those from poorly characterized, non-model organism species. UniProt GO annotations are freely available in a range of formats accessible by both file downloads and web-based views. In addition, the introduction of a new, normalized file format in 2010 has made for easier handling of the complete UniProt-GOA data set.

Interoperability of learning objects copyright in the LUISA semantic learning management system

Semantic Web technology is able to provide the required computational semantics for interoperability of learning resources across different Learning Management Systems (LMS) and Learning Object Repositories (LOR). The EU research project LUISA (Learning Content Management System Using Innovative Semantic Web Services Architecture) addresses the development of a reference semantic architecture for the major challenges in the search, interchange and delivery of learning objects in a service-oriented context. One of the key issues, highlighted in this paper, is Digital Rights Management (DRM) interoperability. A Semantic Web approach to copyright management has been followed, which places a Copyright Ontology as the key component for interoperability among existing DRM systems and other licensing schemes like Creative Commons. Moreover, Semantic Web tools like reasoners, rule engines and semantic queries facilitate the implementation of an interoperable copyright management component in the LUISA architecture.

Content aggregation on knowledge bases using graph clustering

Recently, research projects such as PADLR and SWAP have developed tools like Edutella or Bibster, which are targeted at establishing peer-to-peer knowledge management (P2PKM) systems. In such a system, it is necessary to obtain provide brief semantic descriptions of peers, so that routing algorithms or matchmaking processes can make decisions about which communities peers should belong to, or to which peers a given query should be forwarded. This paper proposes the use of graph clustering techniques on knowledge bases for that purpose. Using this clustering, we can show that our strategy requires up to 58% fewer queries than the baselines to yield full recall in a bibliographic P2PKM scenario.

Semantic network analysis of ontologies

A key argument for modeling knowledge in ontologies is the easy re-use and re-engineering of the knowledge. However, beside consistency checking, current ontology engineering tools provide only basic functionalities for analyzing ontologies. Since ontologies can be considered as (labeled, directed) graphs, graph analysis techniques are a suitable answer for this need. Graph analysis has been performed by sociologists for over 60 years, and resulted in the vivid research area of Social Network Analysis (SNA). While social network structures in general currently receive high attention in the Semantic Web community, there are only very few SNA applications up to now, and virtually none for analyzing the structure of ontologies. We illustrate in this paper the benefits of applying SNA to ontologies and the Semantic Web, and discuss which research topics arise on the edge between the two areas. In particular, we discuss how different notions of centrality describe the core content and structure of an ontology. From the rather simple notion of degree centrality over betweenness centrality to the more complex eigenvector centrality based on Hermitian matrices, we illustrate the insights these measures provide on two ontologies, which are different in purpose, scope, and size.