Istar: um esquema estrela otimizado para Image Data Warehouses baseado em similaridade

Abstract

A data warehousing environment supports the decision-making process through the investigation and analysis of data in an organized and agile way. However, the current data warehousing technologies do not allow that the decision-making processe be carried out based on images pictorial (intrinsic) features. This analysis can not be carried out in a conventional data warehousing because it requires the management of data related to the intrinsic features of the images to perform similarity comparisons. In this work, we propose a new data warehousing environment called iCube to enable the processing of OLAP perceptual similarity queries over images, based on their pictorial (intrinsic) features. Our approach deals with and extends the three main phases of the traditional data warehousing process to allow the use of images as data. For the data integration phase, or ETL phase, we propose a process to represent the image by its intrinsic content (such as color or texture numerical descriptors) and integrate this data with conventional data in the DW. For the dimensional modeling phase, we propose a star schema, called iStar, that stores both the intrinsic and the conventional image data. Moreover, at this stage, our approach models the schema to represent and support the use of different user-defined perceptual layers. For the data analysis phase, we propose an environment in which the OLAP engine uses the image similarity as a query predicate. This environment employs a filter mechanism to speed-up the query execution. The iStar was validated through performance tests for evaluating both the building cost and the cost to process IOLAP queries. The results showed that our approach provided an impressive performance improvement in IOLAP query processing. The performance gain of the iCube over the best related work (i.e. SingleOnion) was up to 98,21%.