Multimedia Retrieval

The SCHEMA Reference System is a content-based image and video indexing and retrieval system that adopts a module-based, expandable architecture. Using this modulebased approach, five different analysis modules, developed at different researcher laboratories, have been integrated with it. Within the system, the MPEG-7 XM (MPEG-7 eXperimentation Model), a non-normative part of the MPEG-7 standard realizing the normative descriptors, is employed along with extensions for descriptor extraction and for supporting search and retrieval functionalities. In addition to the XM, the system supports high level descriptors (e.g. face/non-face image categorization) and content-based indexing and retrieval using other modalities (e.g. pre-existing keyword annotations, text generated via automatic speech recognition (ASR)). In this paper, the TRECVID 2004 test corpus is used as a common data set for demonstrating the functionalities and the efficiency of the proposed system.

In this paper, the most recent version of the system developed by the SCHEMA NoE, termed SCHEMA Reference System, is presented. The Reference System adopts a module-based, expandable architecture, with well defined interfaces between different modules, facilitating the ongoing expansion of the system based on researchers' contributions. The proposed system employs the MPEG-7 XM (MPEG-7 eXperimentation Model) along with extensions developed specifically for the system to improve functionality and efficiency. In addition, the system supports high level descriptors and content-based indexing and retrieval using other modalities (e.g. pre-existing keyword annotations, text generated via automatic speech recognition (ASR)). In this paper, the TREC-VID 2004 test corpus is used as a common data set for demonstrating the functionalities and the efficiency of the proposed system.

This paper presents a system that is designed to make possible the organization and search within the collected digitized material of intangible cultural heritage. The motivation for building the system was a vast quantity of multimedia documents collected by a team from the Institute for Balkan Studies in Belgrade. The main topic of their research were linguistic properties of speeches that are used in various places in the Balkans by different groups of people. This paper deals with a prototype system that enables the annotation of the collected material and its organization into a native XML database through a graphical interface. The system enables the search of the database and the presentation of digitized multimedia documents and spatial as well as non-spatial information of the queried data. The multimedia content can be read, listened to or watched while spatial properties are presented on the graphics that consists of geographic regions in the Balkans. The system also enab...

With the explosive broadcast of multimedia (text documents, image, video etc.) in our life, how to annotate, search, index, browse and relate various forms of information efficiently becomes more and more important. Combining these challenges by relating them to user preference and customization only complicates the matter further. The goal of this survey is to give an overview of the current situation in the branches of research that are involved in annotation, relation and presentation to a user by preference. This paper will present some current models and techniques being researched to model ontology, preference, context, and presentation and bring them together in a chain of ideas that leads from raw uninformed data to an actual usable user interface that adapts with user preference and customization.

With the explosive broadcast of multimedia (text documents, image, video etc.) in our life, how to annotate, search, index, browse and relate various forms of information efficiently becomes more and more important. Combining these challenges by relating them to user preference and customization only complicates the matter further. The goal of this survey is to give an overview of the current situation in the branches of research that are involved in annotation, relation and presentation to a user by preference. This paper will present some current models and techniques being researched to model ontology, preference, context, and presentation and bring them together in a chain of ideas that leads from raw uninformed data to an actual usable user interface that adapts with user preference and customization.

Retrieval of multimedia has become a requirement for many contemporary information systems. These systems need to provide browsing, querying, navigation, and, sometimes, composition capabilities involving various forms of media. In this survey, we review techniques for text, image, audio and video retrieval. We first look at indexing and retrieval techniques for text, audio, image and video. We also discuss features visual features for video retrieval such as colour, texture, shape. The indexing techniques are discussed for these features. We also compare most popular techniques used for indexing and retrieval.

Relevance feedback techniques are designed to automatically improve a system's representation of a query by using documents the user has marked as relevant. However, traditional relevance feedback models suffer from a number of limitations that restrict their potential in supporting information seeking. One of the major limitations of relevance feedback is that it does not incorporate behavioural aspects of information seeking -how and why users assess relevance. We propose that relevance feedback should be viewed as a process of explanation and demonstrate how this limitation of relevance feedback techniques can be overcome by a theory of relevance feedback based on abductive inference.

In our participation in the ImageCLEF 2009 Photo Retrieval task we pursued two objectives: Firstly, to re-evaluate MultiModal Local Context Analysis (MMLCA), our multimodal fusion technique. Secondly, to evaluate a new subquery generation technique based on clustering. From the experiments conducted: Firstly, we confirmed MMLCA performs better for generic domain collections than the other local expansion techniques evaluated. Secondly, our proposal of subquery generation based on clustering, obtained good results (5th best textual run of the Photo Retrieval task). Besides these results in this paper we try to reflect on the extent to which reordering techniques are appropriate to promote diversity. Results suggest that while reordering strategies limit the margin of improvement due to their use of a limited number of documents, the use of approaches based on subqueries generation can overcome this limitation.

Overlay text brings important semantic clues in video content analysis such as video information retrieval and summarization, since the content of the scene or the editors intention can be well represented by using inserted text. Most of the previous approaches to extracting overlay text from videos are based on low-level features, such as edge, color, and texture information. However, existing methods experience difficulties in handling texts with various contrasts or inserted in a complex background. In this paper, we propose a novel framework to detect and extract the overlay text from the video scene. Based on our observation that there exist transient colors between inserted text and its adjacent background, a transition map is first generated. Then candidate regions are extracted by a reshaping method and the overlay text regions are determined based on the occurrence of overlay text in each candidate. In this paper represents comparative approach between edge and change in in...

We propose in this paper a novel multimodal approach to automatically predict the visual concepts of images through an effective fusion of visual and textual features. It relies on a Selective Weighted Late Fusion (SWLF) scheme which, in optimizing an overall Mean interpolated Average Precision (MiAP), learns to automatically select and weight the best experts for each visual concept to be recognized. Experiments were conducted on the MIR Flickr image collection within the ImageCLEF 2011 Photo Annotation challenge. The results have brought to the fore the effectiveness of SWLF as it achieved a MiAP of 43.69 % for the detection of the 99 visual concepts which ranked 2 nd out of the 79 submitted runs, while our new variant of SWLF allows to reach a MiAP of 43.93 %.

Association rule mining (ARM) has been studied in the areas of content-based multimedia retrieval and semantic concept detection due to its high efficiency and accuracy. Two important processes in mining the association rules for classification are rule generation and rule selection. In this paper, a novel high-level feature detection framework using the ARM technique together with the correlations among the feature-value pairs is proposed. A new association rule mining (ARM) algorithm has been developed, where the N -featurevalue pairs are generated using a combined measure based on (1) the existence of the (N -1)-feature-value pairs (where N is larger than 1), (2) the correlation between different Nfeature-value pairs and the concept classes through Multiple Correspondence Analysis (MCA), and (3) the similarity representing the harmonic mean of the inter-similarity and intrasimilarity. The final association classification rules are selected by using the calculated harmonic mean of the similarity values. The proposed framework enables the automatic discovery and generation of the N -feature-value pair association rules from the 1-feature-value pairs for classification. Experimenting with 15 high-level features (concepts) and benchmark data sets from TRECVID, our proposed framework achieves promising performance and outperforms three other wellknown classifiers (Decision Trees, Support Vector Machine, and Neural Networks) which are commonly used for performance comparison in the TRECVID community.

This paper presents an overview and comparative analysis of our systems designed for the TRECVID 2015 [1] multimedia event detection (MED) task. We submitted 17 runs, of which 5 each for the zeroexample, 10-example and 100-example subtasks for the Pre-Specified (PS) event detection and 2 runs for the 10-example subtask for the Ad-Hoc (AH) event detection. We did not participate in the Interactive Run. This year we focus on three different parts of the MED task: 1) extending the size of our concept bank and combining it with improved dense trajectories; 2) exploring strategies for semantic query generation (SQG); and 3) combining our visual classifiers with audio and/or textual classifiers. Among our 17 submitted runs, the following runs achieved top performances: -VIREO MED15 MED15EvalFull PS 0Ex MED p-manualfused 1: zero-example system with manual SQG, fused with textual (OCR) and speech (ASR) information. -VIREO MED15 MED15EvalFull PS 10Ex MED p-ConceptBankIDTEK0OCR 1: 10-example system using our Concept-Bank feature fused with the improved dense trajectories and the 0Ex manual visual system and OCR. -VIREO MED15 MED15EvalFull PS 100Ex MED c-ConceptBankIDTJointProb 1: 100-example system using Concept-Bank feature fused with the improved dense trajectories, using the joint probability to create the score.

The concept of “approximate” searching has applications in a vast number of fields. Some examples are non-traditional databases (eg storing images, fingerprints or audio clips, where the concept of exact search is of no use and we search instead for similar objects), text searching, information retrieval, machine learning and classification, image quantization and compression, computational biology, and function prediction. All those applications have some common characteristics. There is a universe of objects, and a nonnegative ...

This paper introduces the reader to the approach we are taking to develop an ontology that could be used to represent the knowledge inherent in filmed materials. Such an ontology could be used as the semantic basis for multimedia retrieval systems. The proposed approach to ontology development is informed by the earlier work of researchers into folksonomy development and facet analysis. A brief survey of this earlier work is presented before our approach is described.

This paper presents an experimental framework for the Placing tasks, both estimation and verification at MediaEval Benchmarking 2016. The proposed framework provides results for four runs first, using metadata (such as user tags and title of images and videos), second, using visual features extracted from the images (such as tamura), third, by using the textual and visual features together and fourth, using metadata as in the first run but with the training data augmented with external sources. Our work mainly focusses on textual features where we develop a language-based model using bag-of-tags with neighbour based smoothing. The effectiveness of the framework is evaluated through experiments in the placing task.

This paper demonstrates in formal terms that Galilean preinertia is a universal property of all physical objects which allows us to demonstrate, also in formal terms, the impossibility of detecting absolute motion and that motion can only be absolute in nature.

Subspace selection is a powerful tool in data mining. An important subspace method is the Fisher-Rao linear discriminant analysis (LDA), which has been successfully applied in many fields such as biometrics, bioinformatics, and multimedia retrieval. However, LDA has a critical drawback: the projection to a subspace tends to merge those classes that are close together in the original feature space. If the separated classes are sampled from Gaussian distributions, all with identical covariance matrices, then LDA maximizes the mean value of the Kullback-Leibler (KL) divergences between the different classes. We generalize this point of view to obtain a framework for choosing a subspace by 1) generalizing the KL divergence to the Bregman divergence and 2) generalizing the arithmetic mean to a general mean. The framework is named the general averaged divergence analysis (GADA). Under this GADA framework, a geometric mean divergence analysis (GMDA) method based on the geometric mean is studied. A large number of experiments based on synthetic data show that our method significantly outperforms LDA and several representative LDA extensions.

After recalling the Galilean origin of the concept of preinertia, and recalling the concept itself (still unknown to contemporary physics), the article explains the enormous empirical evidence of the concept, as well as the experimental possibility of demonstrating its existence by means of a new scientic instrument, the inertioscope. And although very briey, the formal necessity of its existence is also explained. Finally, the physical relevance of preinertia is considered: it allows to demonstrate the impossibility of detecting absolute motion and the absolute nature of motion.

Embedded Media Barcode Links, or simply EMBLs, are optimally blended iconic barcode marks, printed on paper documents, that signify the existence of multimedia associated with that part of the document content (Figure ). EMBLs are used for multimedia retrieval with a camera phone. Users take a picture of an EMBL-signified document patch using a cell phone, and the multimedia associated with the EMBL-signified document location is displayed on the phone. Unlike a traditional barcode which requires an exclusive space, the EMBL construction algorithm acts as an agent to negotiate with a barcode reader for maximum user and document benefits. Because of this negotiation, EMBLs are optimally blended with content and thus have less interference with the original document layout and can be moved closer to a media associated location. Retrieval of media associated with an EMBL is based on the barcode identification of a captured EMBL. Therefore, EMBL retains nearly all barcode identification advantages, such as accuracy, speed, and scalability. Moreover, EMBL takes advantage of users' knowledge of a traditional barcode. Unlike Embedded Media Maker (EMM) which requires underlying document features for marker identification, EMBL has no requirement for the underlying features. This paper will discuss the procedures for EMBL construction and optimization. It will also give experimental results that strongly support the EMBL construction and optimization ideas.

The aim of this presentation is to review some of the standards, connected with multimedia and their metadata. We start with MPEG family and continue with Open Standards for Interactive TV. Efficient video-streaming is presented. Some standards for audiovisual metadata are outline. We finish with W3C standards.

This paper presents an I/O efficient algorithm for graph pattern matching problem. It is based on decision tree approach proposed by B. T. Messmer and H. Bunke. In that paper, if the time needed for preprocessing is neglected, the computational complexity of their approach is only polynomial in the number of input graph vertices. However, the decision tree is of exponential size. It's not practical for the graphs with large size.. In the new algorithm, we increases the preprocessing time as well as space complexity, it can remarkably reduces the n umber of I/Os and keeps the same time complexity. The algorithm is improved here to reduce its I/O complexity and to achieve a better performance on large graphs.

The use of the join operator in metric spaces leads to what is known as a similarity join, where objects of two datasets are paired if they are somehow similar. We propose an heuristic that solves the 1-NN selfsimilarity join, that is, a similarity join of a dataset with itself, that brings together each element with its nearest neighbor within the same dataset. Solving the problem using a simple brute-force algorithm requires O(n) distance calculations, since it requires to compare every element against all others. We propose a simple divide-and-conquer algorithm that gives an approximated solution for the self-similarity join that computes only O(n 3 2 ) distances. We show how the algorithm can be easily modified in order to improve the precision up to 31% (i.e., the percentage of correctly found 1-NNs) and such that 79% of the results are within the 10-NN, with no significant extra distance computations. We present how the algorithm can be executed in parallel and prove that usin...

The main goal of this paper it is to present our experiments in ImageCLEF 2011 Campaign (Medical Retrieval Task). This edition we use textual and visual information, based on the assumption that the textual module better captures the meaning of a topic. So that, the TBIR module works firstly and acts as a filter, and the CBIR system reorder the textual result list. We also investigate if query expansion with image terms or with modality classification could be a way to improve base queries.

Video is a massive amount of data that contains complex interactions between moving objects. The extraction of knowledge from this type of information creates a demand for video analytics systems that uncover statistical relationships between activities and learn the correspondence between content and labels. However, those are open research problems that have high complexity when multiple actors simultaneously perform activities, videos contain noise, and streaming scenarios are considered. The techniques introduced in this dissertation provide a basis for analyzing video. The primary contributions of this research consist of providing new algorithms for the efficient search of activities in video, scene understanding based on interactions between activities, and the predicting of labels for new scenes.

Users are generally interested in the edge-ranked section of returning search results, according to an analysis of click-through data from a very big search engine log. As a result, search engines must achieve great accuracy with top-ranked documents. While there are many methods for improving video search performance, they either ignore the above factor or have difficulties in practical applications. In this paper, we introduce CR Re-Rating, a flexible and effective re-rating approach for improving recovery efficiency. CR Re-Rating employs a crossreferencing (CR) technique to integrate multimodal data in order to deliver high accuracy for toprated outcomes. Test findings reveal that search quality has greatly improved, particularly for top-ranked results.

Let K be a field of characteristic 2 and G a nonabelian locally finite 2-group. Let V (KG)be the group of units with augmentation 1 in the group algebra KG. An explicit list of groups is given, and it is proved that all involutions in V (KG) commute with each other if and only if G is isomorphic to one of the groups on this list. In particular, this property depends only on G and not at all on K. Let KG be the group algebra of a locally finite p-group G over a field K of characteristic p. Then the normalized unit group An interesting way to study V (KG) is to construct embeddings of important groups into it. D.B. Coleman and D.S. Passman [4] have proved that if G is nonabelian then the wreath product of two cyclic groups of order p is involved in V (KG). Larger wreath products have been constructed by A. Mann and A. Shalev in [7], [8], [9] and . In this paper we answer the question of when dihedral groups cannot be embedded into V (KG). Clearly, only the case p = 2 has to be considered. At the same time we obtain the list of the locally finite 2-groups G such that V (KG) does not contain a subgroup isomorphic to a wreath product of two groups (for the case of odd p see ). We remark, that in the present work the method of proof developed in [3] plays an essential role. Let C 2 n , C 2 ∞ and Q 8 be the cyclic group of order 2 n , the quasicyclic group of type 2 ∞ and the quaternion group of order 8, respectively. An involution is a group element of order 2. For any a, b ∈ G we denote [a, b] = a -1 b -1 ab, a b = b -1 ab.

The indexing and retrieval of multimedia items is difficult due to the semantic gap between the user's perception of the data and the descriptions we can derive automatically from the data using computer vision, speech recognition, and natural language processing. In this contribution we consider the nature of the semantic gap in more detail and show examples of methods that help in limiting the gap. These methods can be automatic, but in general the indexing and retrieval of multimedia items should be a collaborative process between the system and the user. We show how to employ the users' interaction.

This paper deals with multimedia information access. We propose two new approaches for hybrid text-image information processing that can be straightforwardly generalized to the more general multimodal scenario. Both approaches fall in the trans-media pseudo-relevance feedback category. Our first method proposes using a mixture model of the aggregate components, considering them as a single relevance concept. In our second approach, we define trans-media similarities as an aggregation of monomodal similarities between the elements of the aggregate and the new multimodal object. We also introduce the monomodal similarity measures for text and images that serve as basic components for both proposed trans-media similarities. We show how one can frame a large variety of problem in order to address them with the proposed techniques: image annotation or captioning, text illustration and multimedia retrieval and clustering. Finally, we present how these methods can be integrated in two applications: a travel blog assistant system and a tool for browsing the Wikipedia taking into account the multimedia nature of its content.

Abstract. XML processing models and respective languages do not reflect the separation of concerns needed in complex XML applications maintenance. As content complexity grows, there is a need for higher abstraction levels on XML processing composition and modularization. This article presents APP, Architecture for Pipelined Processing, focusing on reusable XML pipelines as the way to achieve separation of concerns in complex XML applications. With APP, execution, configuration and development are fully-separated ...

A fundamental problem in image retrieval is how to improve the text-based retrieval systems, which is known as "bridging the semantic gap". The reliance on visual similarity for judging semantic similarity may be problematic due to the semantic gap between low-level content and higher-level concepts. One way to overcome this problem and increase thus retrieval performance is to consider user feedback in an interactive scenario. In our approach, a user starts a query and is then presented with a set of (hopefully) relevant images; selecting from these images those which are more relevant to her. Then the system refines its results after each iteration, using late fusion methods, and allowing the user to dynamically tune the amount of textual and visual information that will be used to retrieve similar images. We describe how does our approach fit in a real-world setting, discussing also an evaluation of results.

The paper presents the Argos evaluation campaign of video content analysis tools supported by the French Techno-Vision program. This project aims at developing the resources of a benchmark of content analysis methods and algorithms. The paper describes the type of the evaluated tasks, the way the content set has been produced, metrics and tools developed for the evaluations and results obtained at the end of the first phase.

This paper presents an overview of the ImageCLEF 2019 lab, organized as part of the Conference and Labs of the Evaluation Forum-CLEF Labs 2019. ImageCLEF is an ongoing evaluation initiative (started in 2003) that promotes the evaluation of technologies for annotation, indexing and retrieval of visual data with the aim of providing information access to large collections of images in various usage scenarios and domains. In 2019, the 17th edition of ImageCLEF runs four main tasks: (i) a medical task that groups three previous tasks (caption analysis, tuberculosis prediction, and medical visual question answering) with new data, (ii) a lifelog task (videos, images and other sources) about daily activities understanding, retrieval and summarization, (iii) a new security task addressing the problems of automatically identifying forged content and retrieve hidden information, and (iv) a new coral task about segmenting and labeling collections of coral images for 3D modeling. The strong participation, with 235 research groups registering, and 63 submitting over 359 runs, shows an important interest in this benchmark campaign.

In today's world e-learning is one of the popular modes of learning and video lectures are more prominent in keeping learners engaged with course. Internet enabled to keep a large number of video lectures on-line. To search for a required topic or subtopic from this huge video repository is becoming very tedious. One way to search for a particular topic is through keyword based search and it is based on extraction of text content available in lecture video files and to achieve it one has to maintain metadata. To maintain the metadata associated with video the frames of video containing text are required to be processed. As the video contains multiple frames per second it is not required to consider each and every frame. So, the frames containing distinct content called key frames need to be identified. The identification of key frames plays crucial role in the lecture video searching process. In this paper, different techniques for key frame identification are experimentally tested and results were compared.

This work is pertaining to the diversified ranking of web-resources and interconnected documents that rely on a network-like structure, e.g. web-pages. A practical example of this would be a query for the k most relevant web-pages that are also in the same time as dissimilar with each other as possible. Relevance and dissimilarity are quantified using an aggregation of network distance and context similarity. For example, for a specific configuration of the problem, we might be interested in web-pages that are similar with the query in terms of their textual description but distant from each other in terms of the web-graph, e.g. many clicks away. In retrospect, a dearth of work can be found in the literature addressing this problem taking the network structure formed by the document links into consideration. In addition, the vast majority of the approaches that have been proposed in the literature leverage greedy heuristics relying on certain measures that capture only some aspect of the problem. Arguably, this is not enough. In this work, we propose a hill-climbing approach that is seeded with a document collection which is generated using greedy heuristics to diversify initially. More importantly, we tackle the problem in the context of web-pages where there is an underlying network structure connecting the available documents and resources. This is a significant difference to the majority of works that tackle the problem in terms of either content definitions, or the graph structure of the data, but never addressing both aspects simultaneously. To the best of our knowledge, this is the very first effort that can be found to combine both aspects of this important problem in an elegant fashion by also allowing a great degree of flexibility on how to configure the trade-offs of (i) document relevance over resultitems' dissimilarity, and (ii) network distance over content relevance or dissimilarity. Last but not least, we present an extensive evaluation of our methods that demonstrate the effectiveness and efficiency thereof.

We accept this thesis as conforming to the required standard Abstract XML has become a standard format in information exchange and integration. Database support of persistent data storage and query capability is often desired for many XML applications. While it is possible to store XML data in traditional relational databases or object-oriented databases, we also desire high-performance native XML databases that are particularly tailored for XML data. We believe an ideal XML database needs to be generic, automatic, access transparent and performance transparent. With these features as the fundamental success criteria for XML databases, this thesis discusses various existing XML database solutions. With these features as design goals, we develop a native XML database that includes native XML storage and an implementation of the XQuery language. This native database is generic in the sense that it accepts any well-formed XML data, regardless of the actual structure of data, and does n...

Content-based image retrieval systems were introduced as an alternative to avoid the need of manual tagging in traditional keyword-based image retrieval systems. However, the representation of image using visual features only involves a loss of information which is referred to as semantic gap. A number of techniques have been proposed to deal with ‘semantic gap’. This paper reviews existing approaches to handle the well-known ‘semantic gap’ problem in image retrieval systems with a particular focus to approaches based on text and image fusion. Keywords— Image retrieval, Semantic Image Retrieval, CBIR, Text and Image fusion.

As the number of internet users are increasing day by day. So amount of data also increases, so fast response is desired by different users. So most of the researchers are working in relevant information retrival. Proposed work has focus on image retrieval by the combination of textual and visual features. In this work annotations of the image are use for the initial filtering of image from the dataset. After first filtration selected images are sent for visual feature base comparison. On the basis of the visual feature combination final ranking of the images are done as per the input query. Results shows that combination of both type of feature give more effective results as compare to single one.

YOLUM, PINAR. Properties of Referral Networks: Emergence of Authority and Trust (Under the direction of Munindar P. Singh). Developing, maintaining, and disseminating trust in open environments is crucial. We develop a decentralized approach to trust in the context of service location. Service providers and consumers are modeled as autonomous agents participating in a multiagent system that functions as a referral network. When a service is requested, an agent may provide the requested service or give a referral to another agent. The agents can judge the quality of service obtained. Importantly the agents can adaptively select their neighbors, decide with whom to interact, and choose how to give referrals. The agents' actions lead to the evolution of the referral network. We study the emergent properties of referral networks, especially those dealing with their quality, efficiency, and structure. We first show how the exchange of referrals affect locating service providers, then identify undesirable network structures and show under which conditions these network structures emerge. When agents refer and change neighbors in certain specific ways, based on link structure, some agents are identified to be substantially more popular or authoritative than others. These asymmetric distributions of popularity and authoritativeness resemble those seen on

Since 3D models are becoming more popular, the need for effective methods capable of retrieving 3D models are becoming crucial. Current methods require an example 3D model as query. However, in many cases, such a query is not easy to get. An alternative is using a hand-draw sketch as query. We present a structure-based local approach (STELA) for retrieving 3D models using a rough sketch as query. It consists of four steps: get an abstract image, detect keyshapes, compute a local descriptor, and match local descriptors. We represent a 3D model by means of suggestive contours. Our proposal includes an additional step aiming at reducing the number of models that will be compared by our local approach. The proposed method is invariant to position, scale, and rotation changes as well. We evaluate our method using the first-tier precision and compare it with a current global approach (HELO). Our results show an increasing in precision for many classes of 3D models.

In the field of multimedia retrieval in video, text frame classification is essential for text detection, event detection, event boundary detection etc. We propose a new text frame classification method that introduces a combination of wavelet and median-moment with k-means clustering to select probable text blocks among 16 equally sized blocks of a video frame. The same feature combination is used with a new Max-Min Clustering at the pixel level to choose probable dominant text pixels in the selected probable text blocks. For the probable text pixels, a so-called Mutual Nearest Neighbor based Symmetry is explored with a four-quadrant formation centered at the centroid of the probable dominant text pixels to know whether a block is a true text block or not. If a frame produces at least one true text block then it is considered as a text frame otherwise it is a non-text frame. Experimental results on different text and non-text datasets including two public datasets and our own created data show that the proposed method gives promising results in terms of recall and precision at the block and frame levels. Further, we also show how existing text detection methods tend to misclassify non-text frames as text frames in term of recall and precision at both the block and frame levels.

Recommender systems have been systematically applied in industry and academia to help users cope with information uncertainty. However, given the multiplicity of the preferences and needs it has been shown that no approach is suitable for all users in all situations. Thus, it is believed that an effective recommender system should incorporate a variety of techniques and features to offer valuable recommendations and enhance the search experience. In this paper we propose a novel video search interface that employs a multimodal recommender system, which can predict topical relevance. The multimodal recommender accounts for interaction data, contextual information, as well as users' affective responses, and exploits these information channels to provide meaningful recommendations of unseen videos. Our experiment shows that the multimodal interaction feature is a promising way to improve the performance of recommendation.

This paper tries to solve the problem of query generation from multiple media examples, e.g. images, by exploiting a media document representation called feature terms. A feature term denotes a continuous interval of a media feature. This approach (1) helps feature accumulation from multiple examples to generate an efficient query; (2) enables the exploration of text-based retrieval models for multimedia retrieval. Three criteria, minimised χ 2 , minimised AC/DC and maximised entropy, are proposed to optimise feature term selection. Two ranking functions, KL divergence and BM25, are used for relevance estimation. Experiments on Corel photo collection and TRECVid 2006 collection show the effectiveness in image and content-based video retrieval.

Although IP-multicast has been proposed and investigated for years, there are major problems inherent in the IP-multicasting technique, e.g., difficulty to scale up the system, difficulty in allocating a globally unique multicast address, complexity in supporting higher level features such as reliable data transfer and congestion/flow control, more importantly, difficulty to deploy on the current Internet infrastructure due to necessity to change many core routers. Recently, End-System Multicast (ESM) has been proposed as an alternative solution so that multicasting services can be quickly deployed. In this paper, we consider the "architectural" and "optimization" issues on designing an ESM-tree. Specifically, we present a distributed algorithm on how to create and maintain an ESM-tree. We propose a distributed algorithm to perform tree optimization (TO) so that an ESM-tree can dynamically adapt to the changing network condition (e.g., drop in transfer bandwidth) so that the nodes within an ESM-tree can receive data more efficiently. The distributed algorithm has the important theoretical properties that at all times, a tree-topology can be maintained and any node joining, leaving, as well as any tree optimization operation will not "partition" the underlying ESM-tree. Therefore, our work can be used to provide an efficient architectural infrastructure for ESM services. We have implemented a prototype ESM system and carried out experiments to illustrate the effectiveness and the performance gains of our ESM optimization protocol.

Research on multimedia information retrieval (MIR) has recently witnessed a booming interest. A prominent feature of this research trend is its simultaneous but independent materialization within several fields of computer science. The resulting richness of paradigms, methods and systems may, on the long run, result in a fragmentation of efforts and slow down progress. The primary goal of this study is to promote an integration of methods and techniques for MIR by contributing a conceptual model that encompasses in a unified and coherent perspective the many efforts that are being produced under the label of MIR. The model offers a retrieval capability that spans two media, text and images, but also several dimensions: form, content and structure. In this way, it reconciles similarity-based methods with semantics-based ones, providing the guidelines for the design of systems that are able to provide a generalized multimedia retrieval service, in which the existing forms of retrieval...