José Martinez

Fast Incremental Clustering of Gaussian Mixture Speaker Models for Scaling up Retrieval In On-Line Broadcast

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 speakers index, whose complexity becomes logarithmic curve

Selecting a discriminant subset of co-occurrence matrix features for texture-based image retrieval

In the general case, searching for images in a content-based image retrieval (CBIR) system amounts essentially, and unfortunately, to a sequential scan of the whole database. In order to accelerate this process, we want to generate summaries of the image database. In this paper, we focus on the selection of the texture features that will be used as a signature in our forthcoming system. We analysed the descriptors extracted from grey-level co-occurrence matrices’s (COM) under the constraints imposed by database systems.

A flexible image search engine

ndessaig 63 ireste.fr Multimedia searching over Internet has gained substantial popularity in the past two years. Java’s networking features, along with the growing number of Web browsers that can execute Java apple&, facilitate distributed processing. Networking and computational performances are key concerns when considering the use of Java to develop performancesensitive distributed multimedia search engines. This paper describes MISE, the MediaSys Image Search Engine over a large-scale network. After an overview of the architecture, we present the search capabilities of MISE as companion part of image processing.

OTHY: Object To HYpermedia

In this paper, we present a Web-based universal browser for heterogeneous and non federated databases. Recently appeared hypermedia methods are at the core of this system. However, contrary to these methodologies, our tool directly supports the conception, the navigation, and the presentation phases without requiring any modification of the databases. It is mostly based on the OTHY framework, which is a library of classes to develop different kinds of presentation based on well-established hypertext concepts. The design of our tool and a first implementation resulted in a prototype under the O2 OODBMS. This implementation was convincing. Consequently, a Java development and two querying and relational database retro-engineering modules are on the path to be added to this prototype.

Efficient Processing of Nearest Neighbor Queries in Parallel Multimedia Databases

This paper deals with the performance problem of nearest neighbor queries in voluminous multimedia databases. We propose a data allocation method which allows achieving a

Indexing Metric Spaces with Nested Forests

0(\sqrt{n})

A new intersection tree for content-based image retrieval

query processing time in parallel settings. Our proposal is based on the complexity analysis of content based retrieval when it is used a clustering method. We derive a valid range of values for the number of clusters that should be obtained from the database. Then, to efficiently process nearest neighbor queries, we derive the optimal number of nodes to maximize parallel resources. We validated our method through experiments with different high dimensional databases and implemented a query processing algorithm for full knearest neighbors in a shared nothing cluster.

On the usage of clustering for content based image retrieval

Searching for similar objects in a dataset, with respect to a query object and a distance, is a fundamental problem for several applications that use complex data. The main difficulties is to focus the search on as few elements as possible and to further limit the computationally-extensive distance calculations between them. Here, we introduce a forest data structure for indexing and querying such data. The efficiency of our proposal is studied through experiments on real-world datasets and a comparison with previous proposals.

Hierarchical organization of a set of Gaussian mixture speaker models for scaling up indexing and retrieval in audio documents

Retrieval of images based on their contents is a process that requires comparisons of a given query (image) with virtually all the images stored in a database with respect to a given distance function. But this is inapplicable on large databases. The main difficulties and goals are to focus the search on as few images as possible and to further limit the need to compute extensive distances between them. Here, we introduce a variant of a metric tree data structure for indexing and querying such data. Both a sequential and a parallel versions are introduced. The efficiency of our proposal is studied through experiments on real-world datasets.

A Model for Indexing Videos and Still Images from the Moroccan Cultural Heritage

Retrieval of images based on the content is a process that requires the comparison of the multidimensional representation of the contents of a given example with all of those images in the database. To speed up this process, several indexing techniques have been proposed. All of them do efficiently the work up to 30 dimensions [8]. Above that, their performance is affected by the properties of the multidimensional space. Facing this problem, one alternative is to reduce the dimensions of the image representation which however conveys an additional loss of precision. Another approach that has been studied and seems to exhibit good performance is the clustering of the database. On this article we analyze this option from a computational complexity approach and devise a proposal for the number of clusters to obtain from the database, which can lead to sublinear algorithms.