A comparative study of texture descriptor analysis for improving content based image retrieval
Kaouther Zekri, Amel Grissa Touzi, Zied Lachiri · 2017
The selection of the appropriate descriptor for texture analysis in Content Based Image Retrieval (CBIR) has always been a challenging problem in the area of image research. In this paper, we suggest a comparative study of texture analysis methods that allow the selection of suitable combination of descriptors for CBIR. Among the most common descriptors in texture retrieval literature, we find the Co-occurrence matrix, Tamura technique and Log-Gabor filters. These three descriptors are very powerful for CBIR systems. Furthermore, determining the effectiveness of each method at the same time, in the isolation and the combination of several descriptors leads to more accurate results for image retrieval. The proposed techniques are trained and tested for three texture databases. The performance of retrieval is expressed in terms of Precision and Recall and the results showed the superiority of our selected descriptors compared with other proposed method for CBIR in Oracle Database Management System (DBMS).