Uma proposta de construção de índice invertido para recuperação de imagens baseada em conteúdo

Abstract

This work shows an approach to construct an inverted index for Content-Based Image Retrieval. The objective is to speed up the process of querying without loss of quality in the result. To evaluate the effectiveness of the proposal, two systems for retrieval of images were developed. The first system was used as a basis for comparison. It uses techniques for extraction of features and for computing similarity of images usually used in CBIR, namely color moments and Euclidian distance. This approach presents problems such as the poor performance in processing time of large collections of images, such as the Web. To improve performance in the querying process, we proposed a system that indexes the images through a structure often used in textual retrieval, known as inverted index. We also propose the use of a cosine-base distance function to compute the similarity among a given query image and database images. We conducted two tests: the quality of retrieval, and the analysis of the number of arithmetic operations performed by the two systems the calculation of similarity. It is shown significant gain in the number of arithmetic operations reduction without significant loss in quality of retrieved.