| Abstract |
|
This paper presents an image retrieval framework with
scalable image representation and inverted file-based indexing
by incorporating automatically generated visual
keywords. A codebook of visual keywords is implemented
adopting a self-organizing map (SOM)-based vector quantization
on the feature space of segmented image regions. The
codebook is utilized to represent images by calculating the
keyword statistics in the individual images as well as in the
collection as a whole. To reduce the dimensionality of the
sparse feature vector, latent semantic indexing technique is
applied and a similarity matching function is proposed by
exploiting the correlation between visual keywords. A query
expansion strategy is also proposed in the inverted index
based on the topology preserving structure of the SOM. Experimental
results over a collection of 5000 general photographic
images demonstrate the efficiency and effectiveness
of the proposed approach compared to the low-level
histogram-based approaches.
|
 Additional Information
|
Citation:
Md. Mahmudur Rahman, Bipin C. Desai, Prabir Bhattacharya,
"Visual Keyword-based Image Retrieval using Latent Semantic Indexing, Correlation-enhanced Similarity Matching and Query Expansion in Inverted Index,"
ideas,
pp. 201-208,
10th International Database Engineering and Applications Symposium (IDEAS'06),
2006
|
