Um Descritor baseado em análise local de cor para busca de imagens em grandes cole ções

Abstract

Advances in multimedia technology led to a large increase in the number of images digital, in consequence, also grew? ä need for m? ethods and more ef fective cient and to store and retrieve this count? udo multimedia ?? edia. Most m? Proposed ethods alcan in literature? Çam high n ?? íveis of and insufficiency and effi c? CFIA (the top 70% accuracy) however most of them perform experiments using small images bases (less 10,000 images), previously classi fied in good ned categories, thus facilitating search task and consequently increasing ní ?? ble accuracy of the evaluated descriptors. On the other hand, when these m? Ethods are evaluated in large paste? Heterogeneous tions, Ni vel ?? accuracy? and relatively low. Thinking about this problem, this dissertation? Tion proposes the descriptor Location Color Pixel Classi cation (LCPC), an m? Ethod based on local analysis to search from large pictures basis. The proposed approach extracts character ?? color ísticas, classifi ing the pixels as border or inside using the same classi scheme is? tion of m? ethod Border / Interior Pixel Classication (BIC), by? are a simple partitioning scheme, but too much and cient and effi cient to incorporate spatial information about the contents? Udo visual image. Experiments were conducted using three bases of images, including one with more than 100,000 images collected from the Web. The results show that the proposed approach? And much higher when compared with other visual descriptors presented previously in literature, with gains in average accuracy of 51% till is 105%