One person labels one million images.

W2Go a travel guidance system by automatic landmark ranking.

iComics automatic conversion of movie into comics.

Efficient large%2Dscale image annotation by probabilistic collaborative multi%2Dlabel propagation.

Dynamic captioning video accessibility enhancement for hearing impairment.

Movie2Comics a feast of multimedia artwork.

Making computers look the way we look exploiting visual attention for image understanding.

Exploring domain%2Dspecific term weight in archived question search.

Utilizing related samples to learn complex queries in interactive concept%2Dbased video search.

Exploring large scale data for multimedia QA an initial study.

Semantic context modeling with maximal margin Conditional Random Fields for automatic image annotation.

An Eye Fixation Database for Saliency Detection in Images.

A distribution based video representation for human action recognition.

Estimating Poses of World%27s Photos with Geographic Metadata.

Mediapedia Mining Web Knowledge to Construct Multimedia Encyclopedia.

Estimating Poses of World%27s Photos with Geographic Metadata.

Learning Cooking Techniques from YouTube.

VDictionary Automatically Generate Visual Dictionary via Wikimedias.

Video Reference A Video Question Answering Engine.

Prototype hierarchy based clustering for the categorization and navigation of web collections.

Vocabulary Filtering for Term Weighting in Archived Question Search.

Prototype hierarchy based clustering for the categorization and navigation of web collections.

Combining relations for information extraction from free text.

Question Answering over Community%2DContributed Web Videos.

Multimedia Question Answering.
The QACTIS system is being developed for the eventual purpose of providing a user the capability of multilingual question-answering from multimedia. QACTIS was tested at TREC-2005 as a means of identifying its successes and limitations in answering questions specifically from English newswire text as it moves in the direction of multilingual, multimedia question answering. In this paper, we provide a complete overview of those parts of QACTIS which focus specifically on text question-answering, and we analyze the system’s performance at TREC-2005.
Visual query suggestion Towards capturing user intent in internet image search.

Semantic Multimedia
Abstract. Multimedia constitutes an interesting field of application for Semantic Web and Semantic Web reasoning, as the access and management of multimedia content and context depends strongly on the semantic descriptions of both. At the same time, multimedia resources constitute complex objects, the descriptions of which are involved and require the foundation on sound modeling practice in order to represent findings of low- and high level multimedia analysis and to make them accessible via Semantic Web querying of resources. This tutorial aims to provide a red thread through these different issues and to give an outline of where Semantic Web modeling and reasoning needs to further contribute to the area of semantic multimedia for the fruitful interaction between these two fields of computer science. 1 Semantics for Multimedia Multimedia objects are ubiquitous, whether found via web search (e.g., Google 3 or Yahoo! 4 images), or via dedicated sites (e.g., Flickr 5 or YouTube 6)orinthe
Pornprobe an LDA%2DSVM based pornography detection system.

Label to region by bi%2Dlayer sparsity priors.

Scalable detection of partial near%2Dduplicate videos by visual%2Dtemporal consistency.

Video reference question answering on YouTube.

Inferring semantic concepts from community%2Dcontributed images and noisy tags.

Automated localization of affective objects and actions in images via caption text%2Dcum%2Deye gaze analysis.

Video reference question answering on YouTube.

Tour the world a technical demonstration of a web%2Dscale landmark recognition engine.

ViewFocus explore places of interests on Google maps using photos with view direction filtering.

Exploiting internal and external semantics for the clustering of short texts using world knowledge.

NUS%2DWIDE a real%2Dworld web image database from National University of Singapore.

VisionGo towards true interactivity.

A revisit of Generative Model for Automatic Image Annotation using Markov Random Fields.

Tour the world Building a web%2Dscale landmark recognition engine.
Modeling and recognizing landmarks at world-scale is a useful yet challenging task. There exists no readily available list of worldwide landmarks. Obtaining reliable visual models for each landmark can also pose problems, and efficiency is another challenge for such a large scale system. This paper leverages the vast amount of multimedia data on the web, the availability of an Internet image search engine, and advances in object recognition and clustering techniques, to address these issues. First, a comprehensive list of landmarks is mined from two sources: (1) ∼20 million GPS-tagged photos and (2) online tour guide web pages. Candidate images for each landmark are then obtained from photo sharing websites or by querying an image search engine. Second, landmark visual models are built by pruning candidate images using efficient image matching and unsupervised clustering techniques. Finally, the landmarks and their visual models are validated by checking authorship of their member images. The resulting landmark recognition engine incorporates 5312 landmarks from 1259 cities in 144 countries. The experiments demonstrate that the engine can deliver satisfactory recognition performance with high efficiency. 1.
A revisit of Generative Model for Automatic Image Annotation using Markov Random Fields.

Hierarchical spatio%2Dtemporal context modeling for action recognition.

An efficient sparse metric learning in high%2Ddimensional space via l1%2Dpenalized log%2Ddeterminant regularization.

Domain adaptation from multiple sources via auxiliary classifiers.
We propose a multiple source domain adaptation method, referred to as Domain Adaptation Machine (DAM), to learn a robust decision function (referred to as target classifier) for label prediction of patterns from the target domain by leveraging a set of pre-computed classifiers (referred to as auxiliary/source classifiers) independently learned with the labeled patterns from multiple source domains. We introduce a new datadependent regularizer based on smoothness assumption into Least-Squares SVM (LS-SVM), which enforces that the target classifier shares similar decision values with the auxiliary classifiers from relevant source domains on the unlabeled patterns of the target domain. In addition, we employ a sparsity regularizer to learn a sparse target classifier. Comprehensive experiments on the challenging TRECVID 2005 corpus demonstrate that DAM outperforms the existing multiple source domain adaptation methods for video concept detection in terms of effectiveness and efficiency. 1.
Multimedia Evidence Fusion for Video Concept Detection via OWA Operator.

Adaptive Model for Integrating Different Types of Associated Texts for Automated Annotation of Web Images.

A syntactic tree matching approach to finding similar questions in community%2Dbased qa services.

Toward a higher%2Dlevel visual representation for object%2Dbased image retrieval.

Proceedings of the 31st Annual International ACM SIGIR Conference on Research and Development in Information Retrieval
Information retrieval on mind maps – what could it be
Exploring knowledge of sub%2Ddomain in a multi%2Dresolution bootstrapping framework for concept detection in news video.
For TRECVID 2008 concept detection task, we principally focus on: (1) Early fusion of texture, edge and color features TECM, abbreviation of the combined TF*IDF weights based on SIFT features, Edge Histogram, and Color Moments. (2) To improve the training efficiency and explore the knowledge between concepts or hidden sub-domains more easily and efficiently, we propose a novel method based on Latent Dirichlet Allocation (LDA): LDA-based multiple-SVM (LDASVM). We first use LDA to cluster all the keyframes into topics according to the maximum element of the topic-simplex representation vector (TRV) of each keyframe. Then, we train the annotated data in each topic for each concept. During training,
Integrated graph%2Dbased semi%2Dsupervised multiple%2Fsingle instance learning framework for image annotation.

Word2Image towards visual interpreting of words.

Probabilistic optimized ranking for multimedia semantic concept detection via RVM.

Automatic image annotation via local multi%2Dlabel classification.

Adaptive multiple feedback strategies for interactive video search.

VisionGo bridging users and multimedia video retrieval.

Visual Synset Towards a higher%2Dlevel visual representation.

Markovian mixture face recognition with discriminative face alignment.

Pre%2Dattentive discrimination of interestingness in images.
Interestingness is an important aesthetic property, which literally means something that arouses curiosity and is a precursor to attention. Aesthetics is becoming more important as multimedia systems become more human and content centric as opposed to technology centric. In this paper, we use insights from cognitive science, neurophysiology of the early visual system and a mix of human experiments and computational modeling for the purpose of investigating interestingness. Categories in image interestingness and their computational realization are explored through a nontrivial dataset and a real-world problem.
Pre%2Dattentive discrimination of interestingness in images.
Interestingness is an important aesthetic property, which literally means something that arouses curiosity and is a precursor to attention. Aesthetics is becoming more important as multimedia systems become more human and content centric as opposed to technology centric. In this paper, we use insights from cognitive science, neurophysiology of the early visual system and a mix of human experiments and computational modeling for the purpose of investigating interestingness. Categories in image interestingness and their computational realization are explored through a nontrivial dataset and a real-world problem.
Object%2DBased Image Retrieval Beyond Visual Appearances.

Editorial.
calls &amp; calendarEDITORIAL
Advances in Multimedia Modeling
Abstract. Attention is a psychological measurement of human reflection against stimulus. We propose a general framework of highlight detection by comparing attention intensity during the watching of sports videos. Three steps are involved: adaptive selection on salient features, unified attention estimation and highlight identification. Adaptive selection computes feature correlation to decide an optimal set of salient features. Unified estimation combines these features by the technique of multi-resolution auto-regressive (MAR) and thus creates a temporal curve of attention intensity. We rank the intensity of attention to discriminate boundaries of highlights. Such a framework alleviates semantic uncertainty around sport highlights and leads to an efficient and effective highlight detection. The advantages are as follows: (1) the capability of using data at coarse temporal resolutions; (2) the robustness against noise caused by modality asynchronism, perception uncertainty and feature mismatch; (3) the employment of Markovian constrains on content presentation, and (4) multi-resolution estimation on attention intensity, which enables the precise allocation of event boundaries. Key words:highlight detection, attention computation, sports video analysis 1
Advances in Multimedia Modeling
Abstract. Attention is a psychological measurement of human reflection against stimulus. We propose a general framework of highlight detection by comparing attention intensity during the watching of sports videos. Three steps are involved: adaptive selection on salient features, unified attention estimation and highlight identification. Adaptive selection computes feature correlation to decide an optimal set of salient features. Unified estimation combines these features by the technique of multi-resolution auto-regressive (MAR) and thus creates a temporal curve of attention intensity. We rank the intensity of attention to discriminate boundaries of highlights. Such a framework alleviates semantic uncertainty around sport highlights and leads to an efficient and effective highlight detection. The advantages are as follows: (1) the capability of using data at coarse temporal resolutions; (2) the robustness against noise caused by modality asynchronism, perception uncertainty and feature mismatch; (3) the employment of Markovian constrains on content presentation, and (4) multi-resolution estimation on attention intensity, which enables the precise allocation of event boundaries. Key words:highlight detection, attention computation, sports video analysis 1
A Multi%2Dresolution Framework for Information Extraction from Free Text.
Extraction of relations between entities is an important part of Information Extraction on free text. Previous methods are mostly based on statistical correlation and dependency relations between entities. This paper re-examines the problem at the multiresolution layers of phrase, clause and sentence using dependency and discourse relations. Our multi-resolution framework ARE (Anchor and Relation) uses clausal relations in 2 ways: 1) to filter noisy dependency paths; and 2) to increase reliability of dependency path extraction. The resulting system outperforms the previous approaches by 3%, 7%, 4 % on MUC4, MUC6 and ACE RDC domains respectively. 1
Towards the next plateau innovative multimedia research beyond trecvid.
Interactive prototypes are often the best way to convince an audience of a new multimedia technology’s possible impact. Because of its dynamic audiovisual nature, a multimedia application demonstration communicates applied science more effectively than a static description in a journal publication would. Ideally, a multimedia demonstrator grasps the audience’s attention by presenting effective results, advanced multimodal interfaces, novel means of user interaction, or combinations thereof. An interactive demo offers researchers a means to engage their audience in a way that they could never achieve with a written manuscript alone. All major multimedia conferences have adopted demo sessions where researchers, equipped with their laptop-installed system, show their demo to the conference attendees on a one-on-one basis. While effective, these demo sessions lack a common denominator. Individual systems provide solutions for different tasks on different data sets, and the conference attendees have to become familiar with the peculiarities of each individual demo. Moreover, a one-on-one demo session makes it infeasible to present a system to the entire conference audience. In addition, due to their lack of focus, the impact of demo sessions on the audience is suboptimal. Editor’s Note Video search is an experience for the senses. As a result, traditional information retrieval metrics can’t fully measure the quality of a video search system. To provide a more interactive assessment of today’s video search engines, the authors have organized the VideOlympics as a real-time evaluation showcase where systems compete to answer specific video searches in front of a live audience. At VideOlympics, seeing and hearing is believing. —John R. Smith Establishing focus A good way to establish focus is letting several demo systems solve the same problem on a similar data set. A notable example of such a focused effort takes place at the demo session at the National Institute for Standards
The use of topic evolution to help users browse and find answers in news video corpus.

Segregated feedback with performance%2Dbased adaptive sampling for interactive news video retrieval.

Enhancing image annotation by integrating concept ontology and text%2Dbased bayesian learning model.

Annotation of paintings with high%2Dlevel semantic concepts using transductive inference and ontology%2Dbased concept disambiguation.

The use of temporal
We give an efficient procedure for verifying that a finite-state concurrent system meets a specification expressed in a (propositional, branching-time) temporal logic. Our algorithm has complexity linear in both the size of the specification and the size of the global state graph for the concurrent system. We also show how this approach can be adapted to handle fairness. We argue that our technique can provide a practical alternative to manual proof construction or use of a mechanical theorem prover for verifying many finite-state concurrent systems. Experimental results show that state machines with several hundred states can be checked in a matter of seconds.
Use of Generalized Pattern Model for Video Annotation.

Active learning approach to interactive spatio%2Dtemporal news video retrieval.

Interactive Spatio%2DTemporal Visual Map Model for Web Video Retrieval.

News Video Retrieval using Implicit Event Semantics.

Use of Generalized Pattern Model for Video Annotation.

Ontology%2DBased Annotation of Paintings Using Transductive Inference Framework.

Fusion of Region and Image%2DBased Techniques for Automatic Image Annotation.

Auto%2DAnnotation of Paintings Using Social Annotations

Interesting nuggets and their impact on definitional question answering.

TRECVID 2007 Search Tasks by NUS%2DICT.

TRECVID 2007 Search Tasks by NUS%2DICT.

Soft pattern matching models for definitional question answering.
We explore probabilistic lexico-syntactic pattern matching, also known as soft pattern matching, in a definitional question answering system. Most current systems use regular expression-based hard matching patterns to identify definition sentences. Such rigid surface matching often fares poorly when faced with language variations. We propose two soft matching models to address this problem: one based on bigrams and the other on the Profile Hidden Markov Model (PHMM). Both models provide a theoretically sound method to model pattern matching as a probabilistic process that generates token sequences. We demonstrate the effectiveness of the models on definition sentence retrieval for definitional question answering. We show that both models significantly outperform the state-of-the-art manually constructed hard matching patterns on recent TREC data. A critical difference between the two models is that the PHMM has a more complex topology. We experimentally show that the PHMM can handle language variations more effectively but requires more training data to converge. While we evaluate soft pattern models only on definitional question answering, we believe that both models are generic and can be extended to other areas where lexico-syntactic pattern matching can be applied.
Document concept lattice for text understanding and summarization.
We argue that the quality of a summary can be evaluated based on how many concepts in the original document(s) that can be preserved after summarization. Here, a concept refers to an abstract or concrete entity or its action often expressed by diverse terms in text. Summary generation can thus be considered as an optimization problem of selecting a set of sentences with minimal answer loss. In this paper, we propose a document concept lattice that indexes the hierarchy of local topics tied to a set of frequent concepts and the corresponding sentences containing these topics. The local topics will specify the promising sub-spaces related to the selected concepts and sentences. Based on this lattice, the summary is an optimized selection of a set of distinct and salient local topics that lead to maximal coverage of concepts with the given number of sentences. Our summarizer based on the concept lattice has demonstrated competitive performance in Document
Automatically Integrating Heterogeneous Ontologies from Structured Web Pages.

ARE Instance Splitting Strategies for Dependency Relation%2DBased Information Extraction.
Information Extraction (IE) is a fundamental technology for NLP. Previous methods for IE were relying on co-occurrence relations, soft patterns and properties of the target (for example, syntactic role), which result in problems of handling paraphrasing and alignment of instances. Our system ARE (Anchor and Relation) is based on the dependency relation model and tackles these problems by unifying entities according to their dependency relations, which we found to provide more invariant relations between entities in many cases. In order to exploit the complexity and characteristics of relation paths, we further classify the relation paths into the categories of ‘easy’, ‘average ’ and ‘hard’, and utilize different extraction strategies based on the characteristics of those categories. Our extraction method leads to improvement in performance by 3 % and 6 % for MUC4 and MUC6 respectively as compared to the state-of-art IE systems. 1
Transductive inference using multiple experts for brushwork annotation in paintings domain.



Semi%2Dsupervised annotation of brushwork in paintings domain using serial combinations of multiple experts.

Bayesian Learning of Hierarchical Multinomial Mixture Models of Concepts for Automatic Image Annotation.

Video Retrieval Using High Level Features Exploiting Query Matching and Confidence%2DBased Weighting.
Abstract. Recent research in video retrieval has focused on automated, highlevel feature indexing on shots or frames. One important application of such indexing is to support precise video retrieval. We report on extensions of this semantic indexing on news video retrieval. First, we utilize extensive query analysis to relate various high-level features and query terms by matching the textual description and context in a time-dependent manner. Second, we introduce a framework to effectively fuse the relation weights with the detectors ’ confidence scores. This results in individual high level features that are weighted on a per-query basis. Tests on the TRECVID 2005 dataset show that the above two enhancements yield significant improvement in performance over a corresponding state-of-the-art video retrieval baseline. 1
Automatic Person Annotation of Family Photo Album.
Abstract. Digital photographs are replacing tradition films in our daily life and the quantity is exploding. This stimulates the strong need for efficient management tools, in which the annotation of “who ” in each photo is essential. In this paper, we propose an automated method to annotate family photos using evidence from face, body and context information. Face recognition is the first consideration. However, its performance is limited by the uncontrolled condition of family photos. In family album, the same groups of people tend to appear in similar events, in which they tend to wear the same clothes within a short time duration and in nearby places. We could make use of social context information and body information to estimate the probability of the persons ’ presence and identify other examples of the same recognized persons. In our approach, we first use social context information to cluster photos into events. Within each event, the body information is clustered, and then combined with face recognition results using a graphical model. Finally, the clusters with high face recognition confidence and context probabilities are identified as belonging to specific person. Experiments on a photo album containing over 1500 photos demonstrate that our approach is effective. 1
Automatic Indexing and Retrieval of Large Broadcast News Video Collections %2D The TRECVID Experience.

Paraphrase Recognition via Dissimilarity Significance Classification.
We propose a supervised, two-phase framework to address the problem of paraphrase recognition (PR). Unlike most PR systems that focus on sentence similarity, our framework detects dissimilarities between sentences and makes its paraphrase judgment based on the significance of such dissimilarities. The ability to differentiate significant dissimilarities not only reveals what makes two sentences a nonparaphrase, but also helps to recall additional paraphrases that contain extra but insignificant information. Experimental results show that while being accurate at discerning non-paraphrasing dissimilarities, our implemented system is able to achieve higher paraphrase recall (93%), at an overall performance comparable to the alternatives. 1
Morphable Face Reconstruction with Multiple Images.

Face Alignment with Unified Subspace Optimization of Active Statistical Models.

Story Boundary Detection in News Video using Global Rule Induction Technique.

Automatic Indexing and Retrieval of Large Broadcast News Video Collections %2D The TRECVID Experience.


Multi%2Dfaceted contextual model for person identification in news video.
Person identification is very important in the domain of multimedia news as it is often the focus of events in news stories and interest of searchers. However, this detection is impeded by the imprecise audio/visual analysis tools. In this paper, we describe a multimodal and multi-faceted approach to Person-X detection in news video. We make use of multimodal features extracted from text, visual and audio inherent in news video. We also incorporate multiple external sources of news from web and parallel news archives to extract location and temporal profile of the persons. We call this second source of information the multi-faceted context. The multimodal, multi-faceted information is then fused using a RankBoosting approach. Experiments on TRECVID 2003 and 2004 search queries demonstrate that our approach is effective. 1.
Mining dependency relations for query expansion in passage retrieval.
Classical query expansion techniques such as the local context analysis (LCA) make use of term co-occurrence statistics to incorporate additional contextual terms for enhancing passage retrieval. However, relevant contextual terms do not always cooccur frequently with the query terms and vice versa. Hence the use of such methods often brings in noise, which leads to reduced precision. Previous studies have demonstrated the importance of relationship analysis for natural language queries in passage retrieval. However, they found that without query expansion, the performance is not satisfactory for short queries. In this paper, we present two novel query expansion techniques that make use of dependency relation analysis to extract contextual terms and relations from external corpuses. The techniques are used to enhance the performance of density based and relation based passage retrieval frameworks respectively. We compare the performance of the resulting systems with LCA in a density based passage retrieval system (DBS) and a relation based system without any query expansion (RBS) using the factoid questions from the TREC-12 QA task. The results show that in terms of MRR scores, our relation based term expansion method with DBS outperforms the LCA by 9.81%, while our relation expansion method outperforms RBS by 17.49%.
A maximal figure%2Dof%2Dmerit %28MFoM%29%2Dlearning approach to robust classifier design for text categorization.
We propose a maximal figure-of-merit learning (MFoM) approach for robust classifier design, which directly optimizes performance metrics of interest for different target classifiers. The proposed approach, embedding the decision functions of classifiers and performance metrics into the overall training objective, learns the parameters of classifiers in a decision-feedback manner to effectively take into account of both positive and negative training samples, and therefore reduce the required size of the positive training data. It has three desirable properties: (a) it is a performance metric oriented learning; (b) the optimized metric is consistent in both training and evaluation sets; and, (c) it is more robust and less sensitive to data variation, and can handle insufficient training data scenario. We evaluate it on the text categorization task using the Reuters-21578 dataset. Training a F1-based binary tree classifier using MFoM, we observed significantly improved performance and enhanced robustness compared to the baseline and SVM, especially on categories with insufficient training samples. The generality for designing other metric-based classifiers is also demonstrated by comparing the precision, recall, and F1-based classifiers. The results clearly show the consistency in performance
Learning Object Models from Semistructured Web Documents.


Guest editors%27 introduction multimedia modelling.

Proceedings of the 13th ACM International Conference on Multimedia

Image and Video Retrieval

Fusion of Multiple Asynchronous Information Sources for Event Detection in Soccer Video.

Analysis and Retrieval of Paintings Using Artistic Color Concepts.

A Novel Approach to Auto Image Annotation Based on Pairwise Constrained Clustering and Semi%2DNa%EFve Bayesian Model.

Retrieval of News Video Using Video Sequence Matching.

Generic soft pattern models for definitional question answering.
This paper explores probabilistic lexico-syntactic pattern matching, also known as soft pattern matching. While previous methods in soft pattern matching are ad hoc in computing the degree of match, we propose two formal matching models: one based on bigrams and the other on the Profile Hidden Markov Model (PHMM). Both models provide a theoretically sound method to model pattern matching as a probabilistic process that generates token sequences. We demonstrate the effectiveness of these models on definition sentence retrieval for definitional question answering. We show that both models significantly outperform state-of-the-art manually constructed patterns. A critical difference between the two models is that the PHMM technique handles language variations more effectively but requires more training data to converge. We believe that both models can be extended to other areas where lexico-syntactic pattern matching can be applied.
Question answering passage retrieval using dependency relations.
State-of-the-art question answering (QA) systems employ termdensity ranking to retrieve answer passages. Such methods often retrieve incorrect passages as relationships among question terms are not considered. Previous studies attempted to address this problem by matching dependency relations between questions and answers. They used strict matching, which fails when semantically equivalent relationships are phrased differently. We propose fuzzy relation matching based on statistical models. We present two methods for learning relation mapping scores from past QA pairs: one based on mutual information and the other on expectation maximization. Experimental results show that our method significantly outperforms state-of-the-art density-based passage retrieval methods by up to 78 % in mean reciprocal rank. Relation matching also brings about a 50 % improvement in a system enhanced by query expansion.
Dependency relation matching for answer selection.

Using Syntactic and Semantic Relation Analysis in Question Answering.

A semi%2Dna%EFve Bayesian method incorporating clustering with pair%2Dwise constraints for auto image annotation.

Story boundary detection in large broadcast news video archives techniques

Between context%2Daware media capture and multimedia content analysis where do we find the promised land%3F

A bootstrapping framework for annotating and retrieving WWW images.
Most current image retrieval systems and commercial search engines use mainly text annotations to index and retrieve WWW images. This research explores the use of machine learning approaches to automatically annotate WWW images based on a predefined list of concepts by fusing evidences from image contents and their associated HTML text. One major practical limitation of employing supervised machine learning approaches is that for effective learning, a large set of labeled training samples is needed. This is tedious and severely impedes the practical development of effective search techniques for WWW images, which are dynamic and fast-changing. As web-based images possess both intrinsic visual contents and text annotations, they provide a strong basis to bootstrap the learning process by adopting a co-training approach involving classifiers based on two orthogonal set of features – visual and text. The idea of cotraining is to start from a small set of labeled training samples, and successively annotate a larger set of unlabeled samples using the two orthogonal classifiers. We carry out experiments using a set of over 5,000 images acquired from the Web. We explore the use of different combinations of HTML text and visual representations. We find that our bootstrapping approach can achieve a performance comparable to that of the supervised learning approach with an F 1 measure of over 54%. At the same time, it offers the added advantage of requiring only a small initial set of training samples.
An Adaptive Image Content Representation and Segmentation Approach to Automatic Image Annotation.

Representation and retrieval of paintings based on art history concepts.

A hierarchical approach to story segmentation of large broadcast news video corpus.

A MFoM learning approach to robust multiclass multi%2Dlabel text categorization.
We propose a multiclass (MC) classification  approach to text categorization (TC). To fully  take advantage of both positive and negative  training examples, a maximal figure-of-merit  (MFoM) learning algorithm is introduced to train  high performance MC classifiers. In contrast to  conventional binary classification, the proposed  MC scheme assigns a uniform score function to  each category for each given test sample, and  thus the classical Bayes decision rules can now  be applied. Since all the MC MFoM classifiers  are simultaneously trained, we expect them to be  more robust and work better than the binary  MFoM classifiers, which are trained separately  and are known to give the best TC performance.
A Learning%2Dbased Approach for Annotating Large On%2DLine Image Collection.

The fusion of audio%2Dvisual features and external knowledge for event detection in team sports video.

Effectiveness of web page classification on finding list answers.
List question answering (QA) offers a unique challenge in effectively and efficiently locating a complete set of distinct answers from huge corpora or the Web. In TREC-12, the median average F 1 performance of list QA systems was only 6.9%. This paper exploits the wealth of freely available text and link structures on the Web to seek complete answers to list questions. We employ natural language parsing, web page classification and clustering to find reliable list answers. We also study the effectiveness of web page classification on both the recall and uniqueness of answers for web-based list QA.
National University of Singapore at the TREC 13 Question Answering Main Task.

Unsupervised learning of soft patterns for generating definitions from online news.
Breaking news often contains timely definitions and descriptions of current terms, organizations and personalities. We utilize such web sources to construct definitions for such terms. Previous work has identified definitions using hand-crafted rules or supervised learning that constructs rigid, hard text patterns. In contrast, we demonstrate a new approach that uses flexible, soft matching patterns to characterize definition sentences. Our soft patterns are able to effectively accommodate the diversity of definition sentence structure exhibited in news. We use pseudorelevance feedback to automatically label sentences for use in soft pattern generation. The application of our unsupervised method significantly improves baseline systems on both the standardized TREC corpus as well as crawled online news articles by 27 % and 30%, respectively, in terms of F measure. When applied to a state-of-art definition generation system recently fielded in the TREC 2003 definitional question answering task, it improves the performance by 14%.
FADA find all distinct answers.
The wealth of information available on the web makes it an attractive resource for seeking quick answers. While web-based question answering becomes an emerging topic in recent years, the problem of efficiently locating a complete set of distinct answers on the Web is far from being solved. We introduce our system, FADA, which relies on question event analysis, web page clustering, and natural language parsing, to find reliable distinct answers with high recall. The method has been found to be effective in strengthening state-of-the-art Web question answering techniques by emphasizing on answer completeness and uniqueness.
Detecting and Partitioning Data Objects in Complex Web Pages.

A Public Reference Implementation of the RAP Anaphora Resolution Algorithm
This paper describes a standalone, publicly-available implementation of the Resolution of Anaphora Procedure (RAP) given by Lappin and Leass (1994). The RAP algorithm resolves third person pronouns, lexical anaphors, and identifies pleonastic pronouns. Our implementation, JavaRAP, fills a current need in anaphora resolution research by providing a reference implementation that can be benchmarked against current algorithms. The implementation uses the standard, publicly available Charniak (2000) parser as input, and generates a list of anaphora-antecedent pairs as output. Alternately, an in-place annotation or substitution of the anaphors with their antecedents can be produced. Evaluation on the MUC-6 co-reference task shows JavaRAP has an accuracy of 57.9%, similar to the performance given previously in the literature (e.g., Preiss 2002). 1.
Evaluating Keyword Selection Methods for WEBSOM Text Archives.

Global Rule Induction for Information Extraction.

Extracting Key Semantic Terms from Chinese Speech Query for Web Searches.
This paper discusses the challenges and proposes a solution to performing information retrieval on the Web using Chinese natural language speech query. The main contribution of this research is in devising a divide-and-conquer strategy to alleviate the speech recognition errors. It uses the query model to facilitate the extraction of main core semantic string (CSS) from the Chinese natural language speech query. It then breaks the CSS into basic components corresponding to phrases, and uses a multi-tier strategy to map the basic components to known phrases in order to further eliminate the errors. The resulting system has been found to be effective.
A mid%2Dlevel representation framework for semantic sports video analysis.
Sports video has been widely studied due to its tremendous commercial potentials. Despite encouraging results from various specific sports games, it is almost impossible to extend a system for a new sports game because they usually employ different sets of low-level features appropriate for the specific games and closely coupled with the use of game specific rules to detect events or highlights. There is a lack of internal representation and structure to be generic and applicable for many different sports. In this paper, we present a generic mid-level representation framework for semantic sports video analysis. The mid-level representation layer is introduced between the low-level audiovisual processing and high-level semantic analysis. It allows us to separate sports specific knowledge and rules from the low-level and mid-level feature extraction. This makes sports video analysis more efficient, effective, and less ad-hoc for various types of sports. To achieve robustness of the low-level feature analysis, a non-parametric clustering, mean shift procedure, has been successfully applied to both color and motion analysis. The proposed framework has been tested for five field-ball type sports covering duration of about 8 hours. Experiments have shown its robust performance in semantic analysis and event detection. We believe that the proposed mid-level representation framework can be used for event detection, highlight extraction, summarization and personalization of many types of sports video. Categories and Subject Descriptors H.3.1 [Content Analysis and Indexing]: abstracting methods, indexing methods.
VideoQA question answering on news video.
When querying a news video archive, the users are interested in retrieving precise answers in the form of a summary that best answers the query. However, current video retrieval systems, including the search engines on the web, are designed to retrieve documents instead of precise answers. This research explores the use of question answering (QA) techniques to support personalized news video retrieval. Users interact with our system, VideoQA, using short natural language questions with implicit constraints on contents, context, duration, and genre of expected videos. VideoQA returns short precise news video summaries as answers. The main contributions of this research are: (a) the extension of QA technology to support QA in news video; and (b) the use of multi-modal features, including visual, audio, textual, and external resources, to help correct speech recognition errors and to perform precise question answering. The system has been tested on 7 days of news video and has been found to be effective. Categories and Subject Descriptor H.3.1 [Content Analysis and Indexing]: linguistic processing, thesaurus.
VideoQA question answering on news video.
When querying a news video archive, the users are interested in retrieving precise answers in the form of a summary that best answers the query. However, current video retrieval systems, including the search engines on the web, are designed to retrieve documents instead of precise answers. This research explores the use of question answering (QA) techniques to support personalized news video retrieval. Users interact with our system, VideoQA, using short natural language questions with implicit constraints on contents, context, duration, and genre of expected videos. VideoQA returns short precise news video summaries as answers. The main contributions of this research are: (a) the extension of QA technology to support QA in news video; and (b) the use of multi-modal features, including visual, audio, textual, and external resources, to help correct speech recognition errors and to perform precise question answering. The system has been tested on 7 days of news video and has been found to be effective. Categories and Subject Descriptor H.3.1 [Content Analysis and Indexing]: linguistic processing, thesaurus.
Automatic Tracking of Face Sequences in MPEG Video.

A Framework to Customize a Face Model for Reusing Animation.

QUALIFIER Question Answering by Lexical Fabric and External Resources.
One of the major challenges in TRECstyle  question-answering (QA) is to overcome  the mismatch in the lexical representations  in the query space and  document space. This is particularly severe  in QA as exact answers, rather than  docmnents, are required in response to  questions. Most current approaches overcome  the mismatch problem by employing  either data redundancy strategy  through the use of Web or linguistic resources.
Hierarchical Indexing and Flexible Element Retrieval for Structured Document.
Abstract. As more and more structured documents, such as the SGML or XML documents, become available on the Web, there is a growing demand to develop effective structured document retrieval which exploits both content and hierarchical structure of documents and return document elements with appropriate granularity. Previous work on partial retrieval of structured document has limited applications due to the requirement of structured queries and restriction that the document structure cannot be traversed according to queries. In this paper, we put forward a method for flexible element retrieval which can retrieve relevant document elements with arbitrary granularity against natural language queries. The proposed techniques constitute a novel hierarchical index propagation and pruning mechanism and an algorithm of ranking document elements based on the hierarchical index. The experimental results show that our method significantly outperforms other existing methods. Our method also shows robustness to the long-standing problems of text length normalization and threshold setting in structured document retrieval. 1
A Global Rule Induction Approach to Information Extraction.

Fractional scaling of image and video in DCT domain.

A Global Rule Induction Approach to Information Extraction.

ATMRA An Automatic Temporal Multi%2Dresolution Analysis Framework for Shot Boundary Detection.

A bootstrapping approach to annotating large image collection.

A maximal figure%2Dof%2Dmerit learning approach to text categorization.

Structured use of external knowledge for event%2Dbased open domain question answering.
One of the major problems in question answering (QA) is that the queries are either too brief or often do not contain most relevant terms in the target corpus. In order to overcome this problem, our earlier work integrates external knowledge extracted from the Web and WordNet to perform Event-based QA on the TREC-11 task. This paper extends our approach to perform event-based QA by uncovering the structure within the external knowledge. The knowledge structure loosely models different facets of QA events, and is used in conjunction with successive constraint relaxation algorithm to achieve effective QA. Our results obtained on TREC-11 QA corpus indicate that the new approach is more effective and able to attain a confidence-weighted score of above
QUALIFIER In TREC%2D12 QA Main Task.
This paper describes a question answering system and its various modules to solve definition, factoid and list questions defined in the TREC12 Main task. In particular, we tackle the factoid QA task by Event-based Question Answering. Each QA event comprises of elements describing different facets like time, location, object, action etc. By analyzing the external knowledge from pre-retrieved TREC documents, Web documents, WordNet and Ontology to discover the QA event structure, we explore the inherent associations among QA elements and then obtain the answers. There are three subsystems working parallel to handle definition, factoid, and list questions separately. We highlight the shared modules, fine-grained named entity recognition, anaphora resolution and canonicalization co-reference resolution, among the three subsystems as well.
Modeling Web Knowledge for Answering Event%2Dbased Questions.

Querying and Clustering Web Pages about Persons and Organizations.

A cinematic%2Dbased framework for scene boundary detection in video.

A Multi%2DModal Approach to Story Segmentation for News Video.

Learning Pattern Rules for Chinese Named Entity Extraction.

A framework for video scene boundary detection.

An Agent%2Dbased Approach to Chinese Named Entity Recognition.
Chinese NE (Named Entity) recognition is a difficult problem because of the uncertainty in word segmentation and flexibility in language structure. This paper proposes the use of a rationality model in a multi-agent framework to tackle this problem. We employ a greedy strategy and use the NE rationality model to evaluate and detect all possible NEs in the text. We then treat the process of selecting the best possible NEs as a multi-agent negotiation problem. The resulting system is robust and is able to handle different types of NE effectively. Our test on the MET-2 test corpus indicates that our system is able to achieve high F1 values of above 92 % on all NE types. 1.
Retrieving News Stories from a News Integration Archive.

Grouping Web Pages about Persons and Organizations for Information Extraction.
Abstract. Information extraction on the Web permits users to retrieve specific information related to the query especially on the name of a person or organization. As name is non-unique, the same name may be mapped to multiple entities. The aim of this paper is to describe an algorithm to cluster the Web pages returned by the search engine so that pages belonging to different entities are clustered into different group. The algorithm uses named entities as the features to divide the document set into direct or indirect pages. It then uses distinct direct pages as seeds of clusters to group indirect pages into different clusters. The algorithm has been found to be effective for Web-based applications. 1
Temporal Multi%2DResolution Framework for Shot Boundary Detection and Keyframe Extraction.
Video shot boundary detection and keyframe extraction is an important step in many video-processing applications. We observe that video shot boundary is a multi-resolution edge phenomenon in the feature space. In this experiment, we expanded our previous temporal multi-resolution analysis (TMRA) work by introducing the new feature vector based on motion, incorporating functions to detect flash and camera/object motion, and selecting automatic thresholds for noise elimination based on the type of video. The framework is used to extract meaningful keyframes. Experiments show that our new system can detect and characterize both the abrupt (CUT) and gradual (GT) transitions effectively. It has good accuracy for both the detection of transitions as well as their boundaries.
The Integration of Lexical Knowledge and External Resources for Question Answering.
For the short, factoid questions in TREC, the query terms we get from the original questions are either too brief or often do not contain most relevant information in the corpus. It will be very difficult to find the answer (especially exact answer) in a large text document collection because of the gap between the query space and the document space. In order to bridge this gap, there is a need to expand the original queries to include the terms in the document space. In this research, we investigate the integration of both the Web and WordNet in performing local context and lexical correlations to bridge the gap. In order to minimize the noise introduced by the external resources, we explore detailed question classes, fine-grained named entities, and successive constraint relaxation.
The Segmentation and Classification of Story Boundaries in News Video.

Complementary Content.
Technology allows users to interact with a wide variety of information and services. However, more and more users need to integrate complementary content to previously accessed information. Crossmedia applications combine different information pieces, which are stored in different media, as a continuous story. Our study selects the particular case of combining printed material and internet resources for the purpose of delivering complementary information to users. Our investigation conducts a workshop with users composed of a scenario-driven interview and a talk-aloud protocol. This experiment reveals the behavior and difficulties of users when they are combining digital and non-digital media to gather complementary information. In addition, some recommendation is suggested focused on the improvement of the user’s experience in crossmedia application that should be considered by designers.
Comparing Keyword Extraction Techniques for WEBSOM Text Archives.

Guest Editorial Modeling Multimedia Information and Systems.

Stratification Approach to Modeling Video.

Detection of human faces in a compressed domain for video stratification.

Computer Graphics International 2001 %28CGI%2701%29

Building Semantic Perceptron Net for Topic Spotting.
This paper presents an approach to automatically build a semantic perceptron net (SPN) for topic spotting. It uses context at the lower layer to select the exact meaning of key words, and employs a combination of context, co-occurrence statistics and thesaurus to group the distributed but semantically related words within a topic to form basic semantic nodes. The semantic nodes are then used to infer the topic within an input document. Experiments on Reuters 21578 data set demonstrate that SPN is able to capture the semantics of topics, and it performs well on topic spotting task. 1. 
A Match and Tiling Approach to Content%2Dbased Video Retrieval.
This paper presents a novel match-and-tiling approach to retrieve video sequences. The approach considers video similarity matching at two levels -- the shot and sequence levels. At the shot level, we transform the matching of similar shots into a problem of matching video feature trajectories using a longest sub-sequence matching technique. This is to achieve both computational simplicity and retrieval effectiveness. At the sequence level, we view sequence matching as a clustering problem and employ an effective sliding window algorithm to locate multi-occurrences of similar video sequences in the database. The resulting technique is able to retrieve both exact and similar video sequences with different durations and shot ordering. Our results demonstrate that our technique is robust and effective in retrieving similar video sequences.
Automatic proxy%2Dbased watermarking for WWW.

A Replication Strategy for Reducing Wait Time in Video%2DOn%2DDemand Systems.

Detection of text captions in compressed domain video.

Temporal multiresolution analysis for video segmentation.

Video Modeling Using Strata%2DBased Annotation.
ver, setting up the concept structure is a tedious manual process. Stratification represents a middle ground that models video content as overlapping strata of concepts. By judiciously choosing the right level of concepts, the extraction of most strata can be automated. It offers high modeling power to support most complex retrieval and browsing operations. The consideration of what can be automated affects our decisions on what entities should be modeled. We discuss the various types of entities as follows:  Object. Video objects may refer to people in videos, such as sports personalities, an individual like Chairman of the US Federal Reserve Board Alan Greenspan, or a news anchor person. Objects could also refer to buildings, vehicles, or a natural structure like a mountain peak.  Category. The semantic structure of many videos is unique and well structured. For example, news video contents can generally be classified into the categor
Approximating Content%2DBased Object%2DLevel Image Retrieval.

Relevance Feedback Techniques for Image Retrieval Using Multiple Attributes.
This paper proposes a relevance feedback (RF) approach to content-based image retrieval using multiple attributes. The proposed approach has been applied to images&#039; text and color attributes. In  order to ensure that meaningful features are extracted, a pseudo object model based on color coherence vector has been adopted to model color content. The RF approach employs techniques developed in the fields of information retrieval and machine learning to extract pertinent features from these attributes to enhance subsequent retrieval effectiveness. A multiple features RF approach has also been developed that utilizes user&#039;s relevance judgments to estimate the importance of different attributes in integrated content-based image retrieval. The developed system has been tested on a large image collection containing over 12,000 images. The results demonstrate that the proposed RF approaches and pseudo-object based color model are effective.
Color%2DBased Pseudo Object Model for Image Retrieval with Relevance Feedback.

An Empirical Study of Color%2DSpatial Retrieval Techniques for Large Image Databases.
In this paper, we present an experimental evaluation of three color-spatial retrieval techniques — the signaturebased technique, the partition-based algorithm and the cluster-based method. The experimental study is performed on an image database consisting of 12,000 images. With the proliferation of image retrieval mechanisms and the lack of extensive performance study, the experimental results can serve as guidelines in selecting a suitable technique and designing a new technique. 1.
Color%2DBased Relevance Feedback for Image Retrieval.

Relevance Feedback Techniques for Color%2Dbased Image Retrieval.
Color has been widely used in content-based image retrieval systems. The problem with using color is that its representation is low level and hence its retrieval effectiveness is limited. This paper investigates techniques to improving the effectiveness of image retrieval based on colors. It examines the choice of suitable color space and color resolution. It describes two techniques for image retrieval with relevance feedback (RF). The first uses machine learning algorithms to extract significant color  intervals and build the decision tree from the relevant image set to support effective RF. The second employs color coherent vector (CCV), in which the pseudo object information encoded in CCV is used for RF. Both techniques have been tested on a large image database containing over 12,000 images. Tests were also performed to evaluate the effectiveness of retrieval at different color resolutions. The results demonstrate that our RF techniques are effective and a medium color resolution of 176 colors performs the best.
Fast Image Retrieval Using Color%2DSpatial Information.
Abstract. In this paper, we present an image retrieval system that employs both the color and spatial information of images to facilitate the retrieval process. The basic unit used in our technique is a single-colored cluster, which bounds a homogeneous region of that color in an image. Two clusters from two images are similar if they are of the same color and overlap in the image space. The number of clusters that can be extracted from an image can be very large, and it affects the accuracy of retrieval. We study the effect of the number of clusters on retrieval effectiveness to determine an appropriate value for “optimal ” performance. To facilitate efficient retrieval, we also propose a multi-tier indexing mechanism called the Sequenced Multi-Attribute Tree (SMAT). We implemented a two-tier SMAT, where the first layer is used to prune away clusters that are of different colors, while the second layer discriminates clusters of different spatial locality. We conducted an experimental study on an image database consisting of 12,000 images. Our results show the effectiveness of the proposed color-spatial approach, and the efficiency of the proposed indexing mechanism. Key words: Single-colored cluster – Content-based retrieval – Color-spatial information – Sequenced multi-attribute tree 1
On Video%2Don%2DDemand sSrvers with Hierarchical Storage.

Fast Signature%2DBased Color%2DSpatial Image Retrieval.
This paper presents a novel method of retrieving images based on color-spatial information. For each image, its color-spatial information is compactly captured by a set of color signatures and an image signature. During the image retrieval process, the color and image signatures of the query image and those of the database images are compared to determine the set of candidate images to be retrieved. In particulal; our similarity measure takes into account contributions of colors that are perceptually similal; and the weightage on the object colors. To speed up the retrieval process, we also propose a Jiltering mechanism that employs a simple similarity measure to prune away irrelevant images quickly, before computing the more complex similarity measure on potential candidate images. We implemented the proposed method and conducted an experimental study on an image database consisting of 12,000 images. Our results show the retrieval effectiveness and efJiciency of the proposed color-spatial approach. 1 
Disk Striping Strategies for Large Video%2Don%2DDemand Servers.
The storage structure of videos on disks affects the number of concurrent users a video- on-demand system can support and hence the average waiting time. In this paper, we propose a phase-based striping method which has the desirable characteristics that it guarantees the maximum waiting time. We also develop a data replication scheme to further reduce the average waiting time of the phase-based method. These two schemes, together with the conventional sequential striping scheme, are then employed to optimize the waiting time of videos based on their access pattern. Simulations were conducted using a video server containing 36 videos under different loading and hardware configurations. The results show that such optimization can reduce the waiting time significantly. The use of replication method could further improve the server performance by over 25%. The overall results indicate that with limited resources, the use of phase-based striping method with replication is preferable under heavy loads.  Keywords: Video-on-demand, disk striping, replication. 
From Text Description to Animation Sequences.
Most current user manuals for electro-mechanical devices are text-based. The inclusion of interactive 3D graphics and multimedia adds physical dimension and realism to &#034;reading&#034; the manual, and encourages the users to explore and experiment with the manual. The new manual should provide the seamless environment for the users to browse the manual in either text or graphics modes and to navigate freely between the two modes. This paper describes the development of a 3D graphical manual that supports the automatic translation of text descriptions in manual pages to animation. The system uses the frame formalism to model the relationships and dependencies between the graphical components of the device. A concept-based information retrieval (IR) engine is used to perform the text-to-animation translation. In order to eliminate the need to construct a separate concept structure for the IR engine, the system uses the same frame-based graphics model for both the animation and text-to-animation translation. The feasibility of the system is demonstrated using a electronic manual for laser  printer.
Using Domain Knowledge in Querying Image Databases.
Introduction  The advances in computer technology enable us to store and manipulate a massive amount of color images on the computer. Databases containing large on-line image collections are becoming common. Current and potential applications for imaging and other media include: medicine; photo-journalism; marketing and promotion; education and training; and entertainment. The emergence of large and complex image databases has led to the demand for tools that can manage, organize and retrieve images effectively. Most early image retrieval systems use keywords or free-text descriptions supplied by the authors as the basis for retrieval [1, 14]. The problem with text-based retrieval systems is that the text descriptions must be entered manually by the authors, and hence the descriptions tend to be incomplete and inconsistent. Because of this problem, recent systems [5, 6, 8] use  the visual attributes of images, such as colors, textures, and shapes, as indices for retrie
An Integrated Color%2DSpatial Approach to Content%2DBased Image Retrieval.
The use of color information for image retrieval has been used widely in many content-basedretrieval system with some success. However, histogram-based color retrieval techniques su#er from a lack of important spatial knowledge. We discuss a technique of integrating color information with spatial knowledge to obtain an overall impression of the image. The technique involves three steps: the selection of a set of representative colors, the analysis of spatial information of the selectedcolors, and the retrieval process based on the integratedcolor-spatial information. Two color histograms are used to aid in the process of color selection. After deriving the set of representative colors, spatial knowledge of the selectedcolors is obtained using a maximum entropy discretization with event covering method. Aretrieval process is formulated to make use of the spatial knowledge to retrieve relevant images. A prototype image retrieval system has been implemented on the Unix system. It is tested on an image database consisting of 260 images. The result shows substantial improvement over the histogram-basedcolor retrieval methods.
A Video Retrieval and Sequencing System.
Video is an effective medium for capturing the events in the real world around us. There exist a vast amount of video materials covering a wide range of applications. However, wide spread use of video in computer applications is often impeded by the lack of effective tools to manage video information systematically. This paper discusses the design and implementation of a frame-based video retrieval and sequencing system (VRSS). The system is designed to support the entire process of video information management: segmenting, indexing, retrieving and sequencing of video data. A semi-automatic tool is developed to divide video sequence into meaningful shots. Each video shot is logged using text descriptions, audio dialogue and cinematic attributes. A twolayered concept-based model is used as the basis to accurately retrieve relevant video shots based on users&#039; free-text query. A cinematic rule-based virtual editing tool is also developed to sequence the video shots retrieved for presentation within a specified time constraint. The system has been tested on a video documentary on the NUS engineering faculty. The results of video retrieval experiments are encouraging.
Automatic generation and refinement of hypertext links.
this document, rather than simply appearing in it. Phrasier extends this cue by displaying keyphrases in different levels of grey which reflect how representative they are of the viewed document, a ########################    ########################## #######################  ##########################  ############################### #########################  ########  ####  ###########################  ###########  #############  ######  #########  #####  ############  ##############  ###########  ###### ############ ###########  ###  #######  ############  ####################  ###################  ##################################  ######  ######  ####  #################### ################  #####################  ###  #########  ##########  #######################  ##########  ############    #########  ##########  ################# #### ###############  ############ ######### ################# #######  #########################################################################################################################  #####################################################################################################  For any given document, a list of related documents ranked by similarity are retrieved. Links can be created to those documents, constrained by thresholds for similarity values or the number of links required. This supports browsing between nodes that as a whole are related, but does not describe the aspects of the nodes that cause them to be similar. The user can navigate to documents that discuss similar topics, but has little control of what those topics are
Content%2DBased Retrieval of Segmented Images.
this document, we review the evaluation approaches that have been taken in the literature about multimedia retrieval systems. Our review covers far from all related publications, but we believe the sample of papers studied is su#ciently large to draw our (sad) conclusions. 2 Quantitative evaluation
A Concept%2DBased Image Retrieval System.
Current image retrieval techniques, while extremely accurate and effective on a relatively small scale, are unable to scale to very large database conditions due to the limitations imposed by the curse of dimensionality on nearest neighbor searching. In addition to this, the better systems are unsuited to visual browsing of the database. We propose the HUGINN framework for a concept-based image retrieval system, which uses state of the art machine learning techniques to constrain searching and browsing through the use of image concept. This report outlines the current state of the art in image retrieval and outlines a current system using these techniques and a novel technique called probabilistic pose prediction to achieve excellent results for near-duplicate and subimage retrieval. It then details a novel system for boosting machine learners to aid in the extraction of concept from images, then outlines a plan for future research culminating in a complete image retrieval and browsing system which
Supporting Composition in a Hypermedia Environment.

A model for integrating multimedia information around 3D graphics hierarchies.

On the Design of a Frame%2DBased Hypermedia System.

