Databases and Images Group

Content-Based Image Retrieval (CBIR) techniques are based on image feature extractors. However, each individual feature extracted has small contribution to the retrieval process. Therefore, to pose queries able to retrieve useful results, the application software must integrate features from several extractors and specify how to compare the features in detail. This project aims at developing algorithms and techniques to help in preparing the queries embedded in the application software. The quantity of features extracted by a set of feature extractors can be very high, so it is important to have an instrument to select the most relevant ones. The project targets to develop attribute selection techniques aiming specifically at image feature selection. The ground for this research topic are the statistical association rules, fractal-based attribute selection and clustering techniques, as these theoretical tools enable the integration of features obtained automatically (syntactical features) with others identified by human specialists (semantic features), selecting the ones with the highest discrimination ability to be used in the image indexing and retrieval. As the practical basis for this project, it will be employed a PACS (Picture Archiving and Communication System) prototype to store and organize the sets of images. The visualization tools to be developed will help to validate the results, such as the correct identification of clusters and outliers, as directed by each feature extractor. As the clusters of images generated by each feature are distinct, a multi-modal visualization can lead to a better understanding of the information provided by the set of images, and help on decision making. The researchers involved in this project have expertise to develop the proposed techniques, and the integration of the expertise and the results obtained by the researchers involved in this project can contribute to enlarge the related fields.

Research domain:
Multimedia databases, similarity search and retrieval, medical images databases.

Project goals:
Development of algorithms, techniques and tools to knowledge-extraction from large sets of medical images taking advantage of indexing techniques and information visualization approaches. This project aims at gathering the expertise of the research groups involved from the University of São Paulo (Brazil), Bourgogne University at Dijon (France) and San Pablo Catholic University (Peru) to bring new developments to the state of the art on content-based image retrieval techniques, similarity search and image mining algorithms.

Description

Summarizing, this project aims at bringing together the expertise of the researchers involved, as all of them have long worked on the research topics involved in the project. The French group at the University of Bourgogne has worked on providing a multicriteria description model and a computer-based image retrieval systems and query rewriting, developing techniques to integrate human-based description of the images with the automatically-extracted features using regular expressions and its integration with database management systems. The Brazilian group at the University of São Paulo has developed techniques to extract features from medical images and metrics to compare images using those features, techniques to index and retrieve images, and analysis and mining techniques to improve image knowledge discovery. The Peruvian group at the San Pablo Catholic University has developed new image data access methods and employed the Self-Organizing Maps (SOM) to aid in retrieving complex data (such as natural language texts and multidimensional data).

Results

Workshops

Coordinators

France		Brazil		Peru
Richard Chbeir		Agma Juci Machado Traina		Ernesto Cuadros-Vargas
Bourgogne University Laboratoire LE2I - Laboratoire Electronique, Informatique et Image		University of São Paulo ICMC - Instituto de Ciências Matemáticas e de Computação		San Pablo Catholic University
Aile de l'Ingénieur Office GS 16 - BP 4787021078 Dijon CEDEX - France		Av. Trabalhador Sãocarlense, 400. 13566-590 São Carlos, SP - Brazil		Av. Salaverry 301, Vallecito, Arequipa - Peru
Phone: Fax:	+333 80 39 36 55 +333 80 39 68 69	Phone: Fax:	+55 16 3373-9674 +55 16 3373-9751	Phone: Fax:	+51 54 934-1932 +51 54 281517
richard.chbeir at u-bourgogne.fr		agma at icmc.usp.br		ecuadros at spc.org.pe

Participants