Fig 1.
(a) Semantic gap-Corel images of two different semantic categories (i.e. “Mountains” and “Beach”) with close visual appearance; (b) Two sample images of different shapes with close visual and semantic appearance (images used in the figure are similar but not identical to the original images used in the study due to copyright issue, and is therefore for illustrative purposes only).
Fig 2.
Methodology of the BoVW based image representation for CBIR.
Fig 3.
Block diagram of the proposed technique based on visual words fusion (images used in the figure are similar but not identical to the original images used in the study due to copyright issue, and is therefore for illustrative purposes only).
Fig 4.
Sample of images from different semantic categories of the Corel-1000 and Corel-1500 image collections (images used in the figure are similar but not identical to the original images used in the study due to copyright issue, and is therefore for illustrative purposes only).
Table 1.
Performance comparison and statistical analysis of different sizes of the dictionary and features percentages of the image on the Corel-1000 image collection (bold values indicate best performance).
Fig 5.
Performance comparisons of standalone SURF, standalone FREAK, and features fusion of SURF and FREAK descriptors techniques on different sizes of the dictionary for the Corel-1000 image collection.
Fig 6.
Comparison between visual words fusion vs. features fusion of SURF-FREAK using proposed technique on the Corel-1000 image collection.
Table 2.
Performance analysis using adaptive or weighted feature fusion of SURF-FREAK descriptors on the Corel-1000 image collection (bold values indicate best performance).
Table 3.
Experimental details of the reported image collections for the proposed technique.
Fig 7.
Performance comparison in terms of PR-curve on the Corel-1000 image collection.
Table 4.
Semantic category-wise comparative analysis of proposed technique based on visual words fusion with state-of-the-art CBIR techniques by formulating dictionary size of 800 visual words on the Corel-1000 image collection (bold values indicate category-wise best performance).
Table 5.
Semantic category-wise comparative analysis of recall with state-of-the-art CBIR techniques on the Corel-1000 image collection (bold values indicate category-wise best performance).
Fig 8.
Image retrieval result shows a reduction of the semantic gap using automatic image annotation on the semantic category “Flowers” of the Corel-1000 image collection (images used in the figure are similar but not identical to the original images used in the study due to copyright issue, and is therefore for illustrative purposes only).
Fig 9.
Image retrieval result shows a reduction of the semantic gap in the semantic category “Horses” of the Corel-1000 image collection (images used in the figure are similar but not identical to the original images used in the study due to copyright issue, and is therefore for illustrative purposes only).
Fig 10.
Performance comparisons of standalone SURF, standalone FREAK, and features fusion of SURF-FREAK techniques on different sizes of the dictionary for the Corel-1500 image collection.
Fig 11.
MAP performance comparison of the proposed technique based on visual words fusion vs. features fusion of SURF-FREAK technique on different sizes of the dictionary for the Corel-1500 image collection.
Fig 12.
Performance comparison in terms of PR-curve on the Corel-1500 image collection.
Table 6.
MAP and avg. recall comparisons with state-of-the-art CBIR techniques on the Corel-1500 image collection (bold values indicate best performance).
Fig 13.
The retrieved images show a reduction of the semantic gap in response to the query image taken from the semantic category “Tigers” of the Corel-1500 image collection (images used in the figure are similar but not identical to the original images used in the study due to copyright issue, and is therefore for illustrative purposes only).
Fig 14.
Performance comparisons of standalone SURF, standalone FREAK, and features fusion of SURF-FREAK techniques on different sizes of the dictionary for the Caltech-256 image collection.
Fig 15.
MAP performance comparison of the proposed technique based on visual words fusion vs. features fusion of SURF-FREAK techniques on different sizes of the dictionary for the Caltech-256 image collection.
Fig 16.
Performance comparison in terms of PR-curve on the Caltech-256 image collection.
Table 7.
Performance measures comparisons with state-of-the-art CBIR techniques on the Caltech-256 image collection (bold values indicate best performance).
Table 8.
Analysis of the computational complexity (time in seconds) required for feature extraction.
Table 9.
Analysis of the computational complexity (time in seconds) required for query image retrieval (complete framework).