Wavelet Decomposition Feature Parallel Fusion by Quaternion Euclidean Product Distance Matching Score for Finger Texture Verification - Advances in Computational Science and Engineering - page 142

Fig. 2. 2D wavelet decomposition feature parallel fusion workflow

3.1 Finger Texture ROI Extraction

In order to obtain more information from finger texture for wavelet decompo-
sition feature, we define a 12

72 window at the preprocessed finger image, as
our feature ROI (Fig.3(b)). Here the ROI include most area about finger except
fingerprint.
Fig. 3(b) ROI which includes most finger texture feature without fingerprint,
is extracted by the window from the original image with a size of 30

×

90. The
reason we choose the ROI size is that (1) more finger texture information for
wavelets decomposition result in a better recognition performance and (2) the
dimensionality of ROI proposed is suitable for computation of 2D wavelets de-
composition.

×

3.2 Wavelet Decomposition Feature Parallel Fusion

In the algorithm, we obtain partial wavelets coecients as our features. That
is, each quaternion is composed of 4 sub-image coecients as feature of par-
allel fusion, i.e. a group of wavelets coecients of a pixel stand for a quater-
nion, thus a group of quaternions were likely to be acquired for an image,
i.e. {P |P n = c 1 + c 2 i + c 3 j + c 4 k},n =1 , 2 , ···

where n was sequence of pixel
of sub-image by wavelet decomposition. Notice that traditional complex vector
parallel fusion usually faces a common problem, incompatibleness of dimension-
ality of two features. As mentioned in the section 2.3, it is often case that two

Fig. 3. (a) Original image (b) ROI windows