Image Processing Reference
In-Depth Information
A content-based image
retrieval approach based on
document queries
M. Ilie “Dunarea de Jos” University of Galati, Faculty of Automatic Control, Computers, Electrical and Electronics Engineering, Galati, Ro-
This chapter presents a new content-based image retrieval (CBIR) approach, which makes use of
descriptors originating in the local and global search spaces. The algorithm extracts four color
descriptors, one texture descriptor, and two local descriptors which are used to train the corresponding
classifiers based on neural networks. Subsequently, the classifiers are grouped in two weighted majority
voting modules, for local and global characteristics. The system is tested on regular images and on docu-
ment scans obtained from two datasets used a benchmark in previous conferences, in order to verify the
architecture robustness. The experimental results demonstrate the effectiveness of the proposed model.
Neural networks
Image descriptors
Weighted majority voting
The authors would like to thank the Project SOP HRD/107/1.5/S/76822—TOP ACADEMIC of
University “Dunarea de Jos” of Galati, Romania.
1 Introduction
The necessity of the content-based image retrieval (CBIR) phenomenon was imposed by the
problems encountered in different areas. Initially, the image classification was done based on
text labels, which was proven to be very time consuming and error prone. Starting from this
Search WWH ::

Custom Search