Professional Documents
Culture Documents
Loupias 2000
Loupias 2000
Loupias 2000
518
0-7803-6297-7/00/$10.0002000 IEEE
corner-like. To describe different parts of the image, the The wavelet representation gives information about the
set of interesting point should not be clustered in few variations in the signal at different scales. In our retrieval
regions. context, we would like to extract salient points from any
From now on, we will refer to these points as salient part of the image where “something” happens in the signal
points, which are not necessarily corners. We will avoid at any resolution. A high wavelet coefficient (in absolute
the term interest points, which is ambiguous, since it was value) at a coarse resolution corresponds to a region with
previously used in the literature as corner. Wavelet high global variations. The idea is to find a relevant point
representations, which express image variations at to represent this global variation by looking at wavelet
different resolutions, are attractive to extract salient points. coefficients at finer resolutions.
Previous point detectors make use of multiresolution
representation. Chen et al. consider two different Since we use wavelets with a compact support, we know
resolutions to extract comers [3]. In image retrieval from which signal points each wavelet coefficient at the
context, contrast-based points are extracted in [ 2 ] . scale 2’ was computed. We can study the wavelet
However, a lot of points are also extracted in textured coefficients for the same points at the finer scale 2”’.
regions because these regions are contrasted. Points are Indeed there is a set of coefficients at the scale 2’+’
extracted with a specific wavelet in [I] but, since only a computed with the same points as a coefficient W2Jf ( n )
given scale is used, different resolutions features cannot be at the scale 2’ (see [ 7 ] for details). We call this set of
detected.
coefficients the children C(W2,f ( n ) ) of the coefficient
2. FROM WAVELETS TO SALIENT POINTS Wz,f( n ) . The children set in one dimension’ is:
519
leads to an extracted point, but with a lower saliency
value. We then need to threshold the saliency value, in
relation to the -desired number of salient points. We first
obtain the points related to global variations; local
variations also appear if enough salient points are
requested.
(a) 100 Haarsalient points (b) Tracked coefficients Figure 5. Haar salient points examples
520
65 Wavelets are also attractive to extract image features for
60 indexing. These local features would be more related to
55 our salient points.
-
Q 50
6. ACKNOWLEDGMENTS
.- 45
3 40 E. Loupias' guest period in Leiden University was
p' 35 supported by the RhGne-Alpes Region, France
30 (EURODOC grant). Thanks to Dr. D.P. Huijsmans and
25 L--- I Prof. F. Peters, Leiden University, for discussions we had
20 J about this topic.
0 5 10 15 20
Recall (%) 7. REFERENCES
Figure 6. Retrieval results [l] S. Bhattacharjee and T. Ebrahimi, "Image Retrieval Based
We observe that the wavelet-based salient points on Structural Content ", Workshop on Image Analysis for
Multimedia Interactive Services, Heinrich-Hertz-Institut (HHI)
perform better than other detectors for these features and
Berlin, Germany, May 3 1 - June 1 1999.
this database. Daubechies 4 has better performances than [2] S. Bres and J.-M. Jolion, "Detection of Interest Points for
Haar but is computationally more expensive. Random Image Indexation ", 3rd Int. Con$ on Visual Information
points are also used in the experiment: we randomly select Systems, VisuaZ99, Amsterdam, The Netherlands, June 2-4 1999,
points, and compute the Gabor features around these pp. 427-434.
points. Their good result can be explained by their [3] C.-H. Chen, J.-S. Lee and Y.-N. Sun, "Wavelet
spreading in the image. For that reason they lead to a more Transformation for Gray-level Comer Detection ", Pattern
complete representation of the image than some detectors. Recognition, 1995, Vol. 28, No. 6, pp. 853-861.
Obviously, the random points are very unlikely to be [4] I. Daubechies, Orthonormal bases of compactly supported
"
52 I