A universal query mechanism (UQM) for shape-based similarity retrieval in image/video database is proposed. In UQM, we investigate statistically common and salient features among multi-instance query samples to reflect user's definition on shape similarity. For two-dimensional binary shape images, Zernike moments (ZMs) adopted by MPEG-7 are computed as the normal shape descriptors. To get un-biased description by the ZMs, we proposed to locate the minimum bounding circle (MBC) for shape content, which accommodates major shape information in images inside while excludes erroneous noises outside. A fast algorithm to locate the MBC for shapes in image has been developed. The best functionality of the proposed UQM is that, excluding simplicity, new feature sets could be plugged directly into system to improve, while free of degradation in the retrieval performance. Simulations demonstrate that, the UQM-retrieved candidates from thirty thousand images in database really match subjective definitions on shape similarity as the complicated boosting query can do.
[1]
Jun-Wei Hsieh,et al.
Similarity Retrieval in Image Databases by Boosted Common Shape Features Among Query Images
,
2001,
IEEE Pacific Rim Conference on Multimedia.
[2]
MokhtarianFarzin,et al.
A Theory of Multiscale, Curvature-Based Shape Representation for Planar Curves
,
1992
.
[3]
Nikolas P. Galatsanos,et al.
Image retrieval based on similarity learning
,
2000,
Proceedings 2000 International Conference on Image Processing (Cat. No.00CH37101).
[4]
Alireza Khotanzad,et al.
Rotation invariant image recognition using features selected via a systematic method
,
1990,
Pattern Recognit..