Texture Indexing by a Hierarchical Representation

摘要

In this paper, a Hierarchical Entropy based Representation for texture indexing HERTI is presented. The hypothesis is that any texture can be efficaciously represented by means of a 1-D signal obtained by a characteristic curve covering a square (uniform under a given criterion and a given segmentation) region. Starting from such a signal, HER can be then efficaciously applied, taking into account of its generality, for image retrieval by content. Moreover, a Spatial Access Method (SAM), i.e. k-d-Tree, has been utilized in order to improve the search performances. The results obtained on some databases show that HERTI achieves very good performances with few false alarms and dismissals.