Consultas por similaridade em bases de dados complexos utilizando técnica OMNI em SGBDR

Abstract

The necessity of storing evergrowing media in terms of storage size and complexity is a tendency that grew considerably with technology and comunication advances. These data known as complex data demand a much bigger structural storage and analysis complexity when compared with simple data such as words or numbers, besides requiring special queries operators, like Range query (Rq) and k-Nearest Neighbors query (kNNq). Among the complex data set the images are highlighted, which need to be compared using extracted features such as color, shape or texture. This comparison is done by distance calculations between the feature value of the central image of the query in relation to all the remaining images in the database. Query time increases significantly with the increase of database size and complexity. To work around the curse of dimensionality problem, this study proposes the application of the OMNI technique, used to promote a filtering step of the number of images to have their distances calculated, avoiding comparing with the entire database. This technique has additional storage costs for its structures, but can accelerate similarity queries, reducing required execution time.