Digital Signal Processing Reference
In-Depth Information
Fig. 1.
10 trademark images (in Line 1) and the corresponding 10 deformed images for each
4.1
Experiment I
With the increase of matrix order and feature dimension, the descriptive power of the
image will enhance, which, yet, also results in information redundancy and dimension
curse. Therefore, in this experiment, 100 deformed images are set as query images.
Then, in different dimensions, improved and classical distance measure are carried on
respectively to examine the ability of each measure to retrieve original images of these
100 deformed images during the other 1400 images in the database. The retrieval
performance is measured using Accumulative frequency (AF) [15] as follows and
higher value of AF means better retrieval performance.
RULJLQDO
LPDJHV
ZLWKLQ
Q
SRVLWLRQV
$)
Q
u
(12)
1
where N=100 is the number of query images.
The retrieval results of the two measures of Zernike moments (ZMs) and Pseu-
do-Zernike moments (PZMs) are listed in Table 1. The results indicate that, using
Zernike moments as a shape descriptor, the retrieval in 36 dimensions of the feature
