Back

Fast access to very large image databases for CBIR applications
Key words

CBIR, Very large database,  multidimensional indexing.
Abstract 

The main objective of this work is to develop a fast and efficient indexing and searching method of the  K  nearest neighbor which is well adapted to CBIR applications (descriptors with high volume, large dimension, heterogeneous, etc.). On one hand, we want to provide answers to the problems of scalability and the curse of dimensionality and on the other hand, we want to deal with similarity problems that arise in both indexing and CBIR.
 
                                                             

 
We propose two different approaches. The first one uses a multidimensional indexing structure based on approximation approach or filtering, which is an improvement in the RA-Blocks method. It is based on the proposal of an algorithm of subdividing the data space which improves the storage capacity of the index and the CPU times. In the second approach, we propose a multidimensional indexing method suitable for heterogeneous data (color, texture, shape, etc.). It combines a non-linear dimensionality reduction technique with a multidimensional indexing approach based on approximation. This combination allows not only to deal with the curse of dimensionality scalability problems but also to exploit the properties of the non-linear space to find suitable similarity measurement related to the kind of the used data.  Finally,  relevance feedback mechanism based on statistical approach is implemented and the whole system is integrated to IMALBUM. 

This work has been achieved during the PhD thesis of Imane Daoudi

References

"An efficient High-Dimensionnal Indexing Method for Content-Based Image Retrieval in Large Image Databases", I. Daoudi, K. Idrissi, S. Ouatik, A. Baskurt, D.  Aboutajdine, EURASIP JASP: Image communication, 24():775-790, Elsevier. 2009.
"A semi supervised metric learning for content-based image retrieval", I. Daoudi, K. Idrissi, S. EL Alaoui Ouatik. . in Int. Conf. on Signal-Image Technology & Internet-Based Systems. Marrakech 2009.
"Vector Approximation based Indexing for High-Dimensional Multimedia Databases", I. Daoudi, S. Ouatik, A. El Kharraz, K. Idrissi, D. Aboutajdine, in Engineering Letters 16(2):210-218, 2008.
"Kernel Region Approximation Blocks For Indexing Heterogonous Databases".  I. Daoudi, K. Idrissi, S. EL Alaoui Ouatik, in IEEE International Conference on Multimedia & Expo, Hannover - Germany. pp. 1237-1240. 2008.
"Kernel Based Approach for High Dimensional Heterogeneous Image Features Management", I. Daoudi, K. Idrissi, S. EL Alaoui Ouatik,  in CBIR Context. In Advanced Concepts for intelligent Vision Systems (ACIVS). Juan les Pins, France, . 2008.