Image Retrieval from a Forensic Cartridge Case Database - Multimedia Database Retrieval: Technology and Applications - page 156

Table 6.2 Comparison of
scaling parameters for image
registration between real
values of scale parameter (

ˆ

ʱ ˁ m
0.100 0.100 5.3
0.200 0.199 2.5
0.300 0.301 1.8
0.400 0.399 1.2
0.500 0.500 0.9
0.600 0.603 1.1
0.700 0.700 0.5
0.800 0.798 0.5
0.900 0.902 0.6
1.0000 1.0000 0.0
1.100 1.088 2.6
1.200 1.184 3.1
1.300 1.288 1.6
1.400 1.388 1.4
1.500 1.483 2.2
1.600 1.583 1.8
1.700 1.691 0.9
1.800 1.788 1.0
1.900 1.874 2.2
2.000 1.983 1.1
The table also shows the
mean square errors of
image registration (ˁ m )

ʱ

ʱ )

and the estimated values ( ˆ )

obtained by phase correlation

ʱ

visual checking. Specifically, a selected matching area is determined based on its
capability of characterizing the impression marks on the reference image, which
is highly correlated with the corresponding area on the target image at a suitable
alignment position. In addition, the selected area is conveyed as a mark instead of a
background or other normal area, because the mark is usually described by a certain
magnitude of gray-level discontinuity as constituting a significant edge feature.
The adoption of the current ECA method is motivated by the observation of
this human approach to visual checking with regard to the notion of the selected
matching area. To incorporate this criterion into the image matching process, it is
natural to adopt the ECA-extraction method where each selected-matching area is
designated by its corresponding high value of the similarity score using a cross-
covariance coefficient (CCC). This process ignores non-relevant areas where the
CCC value is below a predefined threshold value. Edge-density measurement is also
used to ensure that the selected area contains the features. In view of this, both the
CCC value and the edge density are utilized to select an ECA located locally on the
pair consisting of the reference image and the target image.
Let f

be the target image after registration
to the reference image. In a matching process, the cross-correlation function (CCF)
is usually applied to measure the similarity between the reference image f

[ i , k ] be the reference image, and g [ i , k ]

[ i , k ]