In clustering, consensus clustering aims at providing a single partition fitting a consensus from a set of independently generated. Common procedures, which are mainly statistical and graph-based, are recognized for their robustness and ability to scale-up. In this paper, we provide a complementary and original viewpoint over consensus clustering, by means of algebraic definitions which allow to ascertain the nature of available inferences in a systematic approach (e.g. a knowledge base). We found our approach on the lattice of partitions, for which we shall disclose how some operators can be added with the aim to express a formula representing the consensus. We show that adopting an incremental approach may assist to retain significant amount of aggregate data which fits well with the set of input clustering's. Beyond that ability to model formulae, we also note that its potential cannot be easily captured through such a logical system. It is due to the volatile nature of handling partitions which finally impacts on ability to draw some valuable conclusions.

Distributed data collections are now more and more common due to the emergence of cloud computing, to spatially decentralized businesses, or to the availability of various data sharing web services. Obtain knowledge in such a collection raises then t

Data sharing services on the web host huge amounts of resources supplied and accessed by millions of users around the world. While the classical approach is a central control over the data set, even if this data set is distributed, there is growing interesting in decentralized solutions, because of good properties (in particularity, privacy and scaling up). In this paper, we explore a machine learning side of this work direction. We propose a novel technique for decentralized estimation of probabilistic mixture models, which are among the most versatile generative models for understanding data sets. More precisely, we demonstrate how to estimate a global mixture model from a set of local models. Our approach accommodates dynamic topology and data sources and is statistically robust, i.e. resilient to the presence of unreliable local models. Such outlier models may arise from local data which are outliers, compared to the global trend, or poor mixture estimation. We report experiments on synthetic data and real geo-location data from Flickr.

This paper proposes a solution to the problem of aggregating versatile probabilistic models, namely mixtures of probabilistic principal component analyzers. These models are a powerful generative form for capturing high-dimensional, non Gaussian, data. They simultaneously perform mixture adjustment and dimensionality reduction. We demonstrate how such models may be advantageously aggregated by accessing mixture parameters only, rather than original data. Aggregation is carried out through Bayesian estimation with a specific prior and an original variational scheme. Experimental results illustrate the effectiveness of the proposal.

Today, in many clustering applications we deal with a large amount of data that are delivered in form of data streams. To be able to face the problem of analyzing the data as soon as they are produced, we need to build models that can be incrementally updated. This paper presents an adaptation of a bio-inspired algorithm that dynamically creates and visualizes groups of data, to data stream clustering. We introduce a merge operator that can summarize a group of data and a split operator that uses information of a very small set of supervised data and permits to adapt the clustering to a change in the data stream.

This paper proposes a technique for simplifying a given Gaussian mixture model, i.e. reformulating the density in a more parcimonious manner, if possible (less Gaussian components in the mixture). Numerous applications requiring aggregation of models from various sources, or index structures over sets of mixture models for fast access, may benefit from the technique. Variational Bayesian estimation of mixtures is known to be a powerful technique on punctual data. We derive herein a new version of the variational-Bayes EM algorithm that operates on Gaussian components of a given mixture and suppresses redundancy, if any, while preserving structure of the underlying generative process. A main feature of the present scheme is that it merely resorts to the parameters of the original mixture, ensuring low computational cost. Experimental results are reported on real data.

A distributed statistical estimation technique is presented, with its motivation from, and application to, multimedia content-based indexing. The contribution is a scheme for estimating a multivariate probability density, in the case where this density takes the form of a Gaussian mixture model. They have of broad applicability for multimedia feature modelling. Assuming independently estimated mixtures, we propagate their parameters in a decentralized fashion (gossip) in a network, and aggregate GMMs from connected nodes, to improve estimation. As an improvement through a change of principle over previous work, aggregation is achieved through Bayesian modelling of the GMM component grouping problem and solved using a variational Bayes technique, applied at component level. This determines, through a single, low-cost yet accurate process, assignments of components that should be aggregated and the number of components in the mixture after aggregation. Because only model parameters are exchanged on the network, computational and network load remain very moderate. The scheme is demonstrated on the task of speaker recognition.

This paper deals with pattern recognition in a distributed computing context of the peer-to-peer type, that should be more and more interesting for multimedia data indexing and retrieval. Our goal is estimating of class-conditional probability densities, that take the form of Gaussian mixture models (GMM). Originally, we propagate GMMs in a decentralized fashion (gossip) in a network, and aggregate GMMs from various sources, through a technique that only involves little computation and that makes parsimonious usage of the network resource, as model parameters rather than data are transmitted. The aggregation is based on iterative optimization of an approximation of a KL divergence allowing closed-form computation between mixture models. Experimental results demonstrate the scheme to the case of speaker recognition.

In this paper, we introduce a hierarchical classification approach in the incremental framework of speaker indexing. The technique of incremental generation of speaker-homogeneous segments is applied in the first phase. Then, we propose a hierarchical classification approach that applied in the speaker indexing framework. This approach benefits from the efficiency of Gaussian mixture model (GMM) merge algorithm to the high accuracy of update Gaussian mixture models which referenced by speakers tree index. The adaptive threshold algorithm reduces the cost of exploring the speakers GMM into the balanced binary tree of speaker's index, whose complexity becomes logarithmic curve

We address herein the distributed computation of a probability density estimate. Class-conditional probability density estimation is a central need in multimedia pattern recognition, but has classically be conducted in a centralized fashion. In contrast, the present work is motivated by the perspective of a multimedia indexing and retrieval peer-to-peer system over the Internet. In a decentralized fashion, algorithms and data from various contributors would cooperate towards a collective statistical learning. A typical need is aggregation of probabilistic Gaussian mixture models describing the same class, but estimated on several nodes on different data sets. We tackle this goal through an approach requiring only moderate computation at each node and little data to transit between nodes. Both properties are obtained by fusion models via their (few) parameters, rather than via multimedia data itself. Estimation of the aggregated model is provided by an iterative scheme, derived from a modification on Kullback divergence. We provide experimental results on a speaker recognition task with real data, in a gossip propagation setting.

Usage of mobile devices (phones, digital cameras) raises the need for organizing large personal image collections. In accordance with studies on user needs, we propose a statistical criterion and an associated optimization technique, relying on geo-temporal image metadata, for building and tracking a hierarchical structure on the image collection. In a mixture model framework, particularities of the application and typical data sets are taken into account in the design of the scheme (incrementality, ability to cope with non-Gaussian data, with both small and large samples). Results are reported on real data sets.

The increasing number of satellite and cable television channels is resulting in a soaring number of broadcast programs available to viewers. To alleviate this problem, Personal Video Recorders (multimedia platforms which record TV programs on a hard disk) should integrate a recommender system, which purpose is to filter programs according to their relevance. These systems are based on a user profile, acting as a representative for the user's interests. An important research issue resides in going beyond explicitly user-defined profiles. This paper presents a TV recommender system using fuzzy linguistic summarization technique, which enables automatic learning of the user profile. The logical architecture of the recommender system based on the SAINTETIQ model, as well as the main ideas of the filtering task are introduced in this communication.

This paper addresses a still original issue at the crossroads of multimedia data analysis for content-based retrieval, and wearable computing. As users are acquiring multimedia content personal mobile devices, they are getting also undergoing information overflow. The problem of structuring the content into time-oriented meaningful episodes is addressed, and we argue that geographical location processing is crucial, as a complement to processing audiovisual material. A technique for model-based temporal structuring of one's trajectory during a day is presented, based on a Bayesian/MAP approach, that generates one or several summaries. Experimental results illustrate the applicative interest of the problem addressed and validates the proposed solution.

This paper addresses an original issue at the intersection of image sequence analysis, content-based retrieval and wearable computing. The emerging combination of small-size personal digital imaging and communication device terminals (enhanced mobile phones etc.) is enabling the build up of large personal image collections, which induce content-based retrieval issues particular to this context. We consider here the case of a worn camera, automatically and regularly taking pictures, so as to serve as a visual memory for its user. We propose a technique based on face detection which assists this user in finding, in such "slow video" collections, time intervals corresponding to meetings with people. To this purpose, face detection is first run independently on successive images. Its noisy output is then considered the observation sequence in a regularization process conducted with a Viterbi estimation algorithm. The result can be usefully overlaid on a PDA calendar manager. Experiments validate the technique on real data.

We aim at detecting moving objects in color image sequences acquired with a mobile camera. This issue is of key importance in many application fields. To accurately recover motion boundaries, we exploit a fine spatial image partition supplied by a MRF-based color segmentation algorithm. We introduce a region-level graph modeling embedded in a Markovian framework to detect moving objects in the scene viewed by a mobile camera. This is stated as the binary segmentation into regions conforming or not conforming to the dominant image motion assumed to be due to the camera movement. The method is validated on real image sequences.

This paper describes an original approach to partitioning of a video document into shots. Instead of an interframe similarity measure which is directly intensity based, we exploit image motion information, which is generally more intrinsic to the video structure itself. The proposed scheme aims at detecting all types of transitions between shots using a single technique and the same parameter set, rather than a set of dedicated methods. The proposed shot change detection method is related to the computation, at each time instant, of the dominant image motion represented by a two-dimensional affine model. More precisely, we analyze the temporal evolution of the size of the support associated to the estimated dominant motion. Besides, the computation of the global motion model supplies by-products, such as qualitative camera motion description, which we describe in this paper, and other possible extensions, such as mosaicing and mobile zone detection. Results on videos of various content types are reported and validate the proposed approach.

This paper presents the first version of our platform for automatically building the structure of a video sequence. The first application uses semi-automatic tools based only on image analysis for building interactive videos: decomposing the video into shots, extracting and tracking objects within each shot and linking occurrences of similar objects among the shots. The second application provides the end user with a powerful browser to navigate through any preprocessed hypervideo.

This paper deals with the problem of motion-based segmentation of image sequences. Such partitions are multiple-purpose in dynamic scene analysis. We first extract a spatial texture-based partition using an unsupervised MRF approach. The regions obtained are then grouped according to a motion-based criterion. This grouping process relies on two motion estimation techniques and exploits centextual information between regions. In contrast with clustering techniques, region grouping is formalized as a motion-based graph labeling process, within a Markovian framework. Results on real-world image sequences are shown and validate the proposed method.