Um sistema para recuperação de imagens com base em características geométricas, conjuntos nebulosos e indexação métrica

Abstract

In this work, the MIGFIR (Metric Indexing and Geometric Features-based Image Retrieval) system is presented. This system was developed mainly to make possible to index and retrieve certain image classes (like human face images, for example) with efficiency and accuracy using geometric measures, two new similarity functions and the metric access method called Slim-tree. This system also allows configuring important parameters during the definition of queries and makes possible to apply fuzzy sets and related concepts so that the language used in search expressions of one of the query classes that are supported be considerably close of the language usually applied by human beings. Efficiency and accuracy of the system at the execution of queries on several real and synthetic human face image sets are also presented. About efficiency, the obtained results show that MIGFIR can promote a quite considerable reduction of the execution time of queries in comparison with a similar approach that applies sequential scanning of the data. About accuracy, the results show that performance is satisfactory during the retrieval of human faces similar to certain example-images.