We present a general strategy for shape-based image retrieval which considers similarity modulo a given transformation group G. The shape content of an image is summarized by recording what geometric primitives, such as line segments and circular arcs, t where in the image. Geometric hashing is used to compute a set of primitive features which are invariant under a G-transformation of the image. Our search engine is feature-based in the sense that similarity is determined by looping over the features in the query and asking: Which database images have features that are close to a given query feature? The most similar database images are ones that have many features which are close to query features. We apply our approach to an example database of 500 chinese character bitmaps.
[1]
Geoff A. W. West,et al.
Nonparametric Segmentation of Curves into Various Representations
,
1995,
IEEE Trans. Pattern Anal. Mach. Intell..
[2]
Sergey Brin,et al.
Near Neighbor Search in Large Metric Spaces
,
1995,
VLDB.
[3]
Yehezkel Lamdan,et al.
Geometric Hashing: A General And Efficient Model-based Recognition Scheme
,
1988,
[1988 Proceedings] Second International Conference on Computer Vision.
[4]
Leonidas J. Guibas,et al.
Partial matching of planar polylines under similarity transformations
,
1997,
SODA '97.
[5]
David G. Lowe,et al.
Three-Dimensional Object Recognition from Single Two-Dimensional Images
,
1987,
Artif. Intell..