Recuperación de imágenes mediante rasgos descriptores globales y locales

Abstract

Feature extraction is a key issue in Content Based Image Retrieval (CBIR). In the past, a number of textures have been proposed in literature, including statistic methods. In this work is proposed an extraction and features classification methodology, applied to scenes retrieval of natural images. The proposed Methodology uses random points which are input to a 1-nn classifier with the purpose of testing how discriminating the mean, standard deviation are and homogeneity (from a co-occurrence matrix) features combination to describe different classes in a scene. It also proposes that the use of-K-means algorithm to find unsupervised groups or clusters (uncorrelated) that exist in a natural scene and the objects in scene are not associated with the labels normally a user makes an image to describe the contents.