This thesis presents work aimed at exploiting the grammar of film for the purpose of automated film understanding, and addresses the semantic gap that exists between the simplicity of features that can be currently computed in automated content indexing systems and the richness of semantics in user queries posed for media search and retrieval. The problem is set within the broader context of the need for enabling technologies for multimedia content management, and arises in response to the growing presence of multimedia data made possible by advances in storage, processing, and transmission technologies. The first demonstration of this philosophy uses the attributes of motion and shot length to define and compute a novel measure of film tempo. Tempo flow plots are defined and derived for a number of full length movies, and edge analysis is performed leading to the extraction of dramatic story sections and events signaled by their unique tempo. In addition to the development of this computable tempo measure, a study is conducted as to the usefulness of biasing it toward either of its constituents, namely motion or shot length. Thirdly, a refinement is made to the shot length normalizing mechanism, driven by the peculiar characteristics of shot length distribution exhibited by movies. The next aspect of film examined is film rhythm. In the rhythm model presented, motion behaviour is classified as being either nonexistent, fluid or staccato for a given shot. Shot neighbourhoods in movies are then grouped by proportional makeup of these motion behavioural classes to yield seven high-level rhythmic arrangements that prove adept at indicating likely scene content (e.g., dialogue or chase sequence). The second part of the investigation presents a novel computational model to detect editing patterns as either metric, accelerated, decelerated, or free. / It is also found that combined motion and editing rhythms allow us to determine that the media content has changed and hypothesize as to why this is so. Three such categories are presented along with their efficacy for capturing useful film elements (e.g., scene change precipitated by plot event). Finally, the first attempt to extract narrative structure, the prevalent 3-Act storytelling paradigm in film, is detailed. The identification of act boundaries in the narrative allows for structuralizing film at a level far higher than existing segmentation frameworks which include shot detection and scene identification, and provides a reliable basis for inferences about the semantic content of dramatic events in film. Additionally, the narrative constructs identified have analogues in many other domains, including news, training video, sitcoms, etc., making these ideas widely applicable. A novel act boundary posterior function for Act 1 and 2 is derived using a Bayesian formulation under guidance from film grammar, tested under many configurations, and the results are reported for experiments involving 25 full-length movies. The framework is shown to have a role in both the automatic and semi-interactive setting for semantic analysis of film.

All practical text search systems use inverted indexes to quickly resolve user queries. Offline index construction algorithms, where queries are not accepted during construction, have been the subject of much prior research. As a result, current techniques can invert virtually unlimited amounts of text in limited main memory, making efficient use of both time and disk space. However, these algorithms assume that the collection does not change during the use of the index. This thesis examines the task of index maintenance, the problem of adapting an inverted index to reflect changes in the collection it describes. Existing approaches to index maintenance are discussed, including proposed optimisations. We present analysis and empirical evidence suggesting that existing maintenance algorithms either scale poorly to large collections, or significantly degrade query resolution speed. In addition, we propose a new strategy for index maintenance that trades a strictly controlled amount of querying efficiency for greatly increased maintenance speed and scalability. Analysis and empirical results are presented that show that this new algorithm is a useful trade-off between indexing and querying efficiency. In scenarios described in Chapter 7, the use of the new maintenance algorithm reduces the time required to construct an index to under one sixth of the time taken by algorithms that maintain contiguous inverted lists. In addition to work on index maintenance, we present a new technique for accumulator pruning during ranked query evaluation, as well as providing evidence that existing approaches are unsatisfactory for collections of large size. Accumulator pruning is a key problem in both querying efficiency and overall text search system efficiency. Existing approaches either fail to bound the memory footprint required for query evaluation, or suffer loss of retrieval accuracy. In contrast, the new pruning algorithm can be used to limit the memory footprint of ranked query evaluation, and in our experiments gives retrieval accuracy not worse than previous alternatives. The results presented in this thesis are validated with robust experiments, which utilise collections of significant size, containing real data, and tested using appropriate numbers of real queries. The techniques presented in this thesis allow information retrieval applications to efficiently index and search changing collections, a task that has been historically problematic.

This thesis presents a research work based on an integrated multi-modal approach for sports video indexing and retrieval. By combining specific features extractable from multiple (audio-visual) modalities, generic structure and specific events can be detected and classified. During browsing and retrieval, users will benefit from the integration of high-level semantic and some descriptive mid-level features such as whistle and close-up view of player(s). The main objective is to contribute to the three major components of sports video indexing systems. The first component is a set of powerful techniques to extract audio-visual features and semantic contents automatically. The main purposes are to reduce manual annotations and to summarize the lengthy contents into a compact, meaningful and more enjoyable presentation. The second component is an expressive and flexible indexing technique that supports gradual index construction. Indexing scheme is essential to determine the methods by which users can access a video database. The third and last component is a query language that can generate dynamic video summaries for smart browsing and support user-oriented retrievals.

<p>Mängden information på internet är enorm och bara forsätter att öka på både gott och ont. Framförallt kan det vara svårt för grupper såsom synskadade och personer med språksvårigheter att navigera sig och ta vara på all denna information. Därmed finns ett behov av väl fungerande sammanfattningsverktyg för dessa, men även för andra människor som snabbt behöver presenteras det viktigaste ur en uppsättning texter. Den här studien undersöker hur väl sammanfattningssystemet CogSum, som är baserat på Random Indexing, presterar med och utan rankningsalgoritmen PageRank aktiverat på nyhetstexter och texter från Försäkringskassan. Utöver detta används sammanfattningssystemet SweSum som en baslinje i undersökningen. Rapporten innefattar en teoretisk bakgrund som avhandlar automatisk textsammanfattning i stort vilket inkluderar olika utvärderingsmetoder, tekniker och sammanfattningssystem. Utvärderingen utfördes med hjälp av det automatiska utvärderingsverktyget KTHxc på nyhetstexterna och ett annat sådant, AutoSummENG, på Försäkringskassans texter. Studiens resultat påvisar att CogSum utan PageRank presterar bättre än CogSum med PageRank på 10 nyhetstexter medan det omvända gäller för 5 texter från Försäkringskassan. SweSum i sin tur erhöll det bästa resultatet för nyhetstexterna respektive det sämsta för texterna från Försäkringskassan.</p>

Most existing Peer-to-Peer (P2P) systems support only title-based searches and are limited in functionality when compared to today’s search engines. In this paper, we present the design of a distributed P2P information sharing system that supports semantic-based content searches of relevant documents. First, we propose a general and extensible framework for searching similar documents in P2P network. The framework is based on the novel concept of Hierarchical Summary Structure. Second, based on the framework, we develop our efficient document searching system, by effectively summarizing and maintaining all documents within the network with different granularity. Finally, an experimental study is conducted on a real P2P prototype, and a large-scale network is further simulated. The results show the effectiveness, efficiency and scalability of the proposed system. / Singapore-MIT Alliance (SMA)

We show that we can optimally represent the set of 2D images produced by the point features of a rigid 3D model as two lines in two high-dimensional spaces. We then decribe a working recognition system in which we represent these spaces discretely in a hash table. We can access this table at run time to find all the groups of model features that could match a group of image features, accounting for the effects of sensing error. We also use this representation of a model's images to demonstrate significant new limitations of two other approaches to recognition: invariants, and non- accidental properties.

This paper describes a new approach to the model base indexing stage of visual object recognition. Fast model base indexing of 3D objects is achieved by accessing a database of encoded 2D views of the objects using a fast 2D matching algorithm. The algorithm is specifically intended as a plausible solution for the problem of indexing into very large model bases that general purpose vision systems and robots will have to deal with in the future. Other properties that make the indexing algorithm attractive are that it can take advantage of most geometric and non-geometric properties of features without modification, and that it addresses the incremental model acquisition problem for 3D objects.

We discuss a strategy for visual recognition by forming groups of salient image features, and then using these groups to index into a data base to find all of the matching groups of model features. We discuss the most space efficient possible method of representing 3-D models for indexing from 2-D data, and show how to account for sensing error when indexing. We also present a convex grouping method that is robust and efficient, both theoretically and in practice. Finally, we combine these modules into a complete recognition system, and test its performance on many real images.

Modern context-aware applications no longer justify their decisions based only on their own information but on the decisions and information of other applications in a similar context.  Acquiring context information of other entities in an distributed system is difficult task when using the current content centric solutions such as DHTs.  This project aims to build a distributed index that provides storage for the so called Presentities solely based on the state of their context information.  Furthermore, the stored Presentities must be efficiently accessible even if only some information of their current context is available. To fulfill these requirements the PAST DHT was extended to support range queries and modified to use points on a space-filling curve as index values. The simulation of the system has shown very good accuracy rates, on average 99%, for range queries by maintaining a logarithmic relationship to the amount of required messages sent in the DHT.  Problems have emerged from the lack of load balancing implemented into the used DHT, but it is still the case that the proposed method of using space-filling curves to build a context centric decentralized index is both sufficient and effective. Keywords: context awareness, indexing, space-flling curves, Hilbert curve,Pastry, PAST

Indexing data for efficient search capabilities is a core problem in many domains of computer science. As applications centered around semantic data sources become more common, the need for more sophisticated indexing and querying capabilities arises. In particular, the need to search for specific information in the presence of a terminology or ontology (i.e. a set of logic based rules that describe concepts and their relations) becomes of particular importance, as the information a user seeks may exists as an entailment of the explicit data by means of the terminology. This variant on traditional indexing and search problems forms the foundation of a range of possible technologies for semantic data. In this work, we propose an ordering language for specifying partial orders over semantic data items modeled as descriptions in a description logic. We then show how these orderings can be used as the basis of a search tree index for processing \emph{concept searches} in the presence of a terminology. We study in detail the properties of the orderings and the associated index structure, and also explore a relationship between ordering descriptions called \emph{order refinement}. A sound and complete procedure for deciding refinement is given. We also empirically evaluate a prototype implementation of our index structure, validating its potential efficacy in semantic query problems.