Figure 3 shows example of salient points and harris points. Harris points
concentrated on the woman's lace area while salient points are more spread in the image.
Figure 3. Corner vs. Salient Points
(left) Harris (right) Salient
Figiure 4 shows two examples of the retrieved images in MARS using the salient points.
The top left image is the query. The query images are airplane and tiger, respectively.
The top 20 retrieved images are shown in Figure 4(a) and 4(c). The corresponding salient points
together with their 3x3 neighborhood are shown in Figure 4(b) and 4(d), respectively.
In the example of airplane, its background is relatively simpler (smaller variations) than the background of
tiger and thus fewer points are located at the background compared to the tiger example. Therefore satisfied retrieval results
are obtained (Fig. 4 (a) and (b)) while in the example of tiger, the flowers (wrong classifications) are retrieved, i.e., the
16th and 17th retrieved images. The reason is that the flowers and tiger have a similar color information, i.e., yellow, in the
neighborhood of salient points.
(a) Query image: airplane (top-left image)
(b) Salient points (50 points) and their 3x3 neighborhood
(c) Query image: tiger (top-left image)
(b) Salient points (50 points) and their 3x3 neighborhood
Figure 4. User interface: experimental results using the color moments extracted from the
3x3 neighborhood of the salient points (rank from left to right, from top to bottom, the top left
is the query image).
Problem Statement
