Wavelets 447
The vectors contain the same values, but in different orders and with different signs. As for
the Haar wavelet, we can apply the Daubechies-4 wavelet by a matrix multiplication; the
matrix f or a 1-scale DWT on a vector of length 8 is
h
0
h
1
h
2
h
3
0 0 0 0
0 0 h
0
h
1
h
2
h
3
0 0
0 0 0 0 h
0
h
1
h
2
h
3
h
2
h
3
0 0 0 0 h
0
h
1
h
3
−h
2
h
1
−h
0
0 0 0 0
0 0 h
3
−h
2
h
1
−h
0
0 0
0 0 0 0 h
3
−h
2
h
1
−h
0
h
1
−h
0
0 0 0 0 h
3
−h
2
Notice that the filter coefficients overlap between rows, which is not the case for the Haar
matrix H
2
n
. This means that the use of the Daubechies-4 wavelet will have smoother results
than using the Haar wavelet. The form of the matrix above is similar to circular convolution
with the one-dimensional filters
h
0
h
1
h
2
h
3
and
h
3
−h
2
h
1
−h
0
.
The discrete wavelet transform can indeed be approached in terms of filtering; the above
two filters are then known as the low pass and high pass filters, respectively. Steps for
performing a 1-scale wavelet transform are given by Umbaugh [52]:
1. Convolve the image rows with the low pass filter.
2. Convolve the columns of the result of Step 1 with the low pass filter, and rescale this
to half its size by subsampling.
3. Convolve the result of Step 1 with the high pass filter, and again subsample to obtain
an image of half the size.
4. Convolve the original image rows with the high pass filter.
5. Convolve the columns of the result of Step 4 with the low pass filter, and rescale this
to half its size by subsampling.
6. Convolve the result of Step 4 with the high pass filter, and again subsample to obtain
an image of half the size.
At the end of these steps there are four images, each half the size of the original. They are:
1. the “low pass/low pass” image (LL); result of Step 2.
2. the “low pass/high pass” image (LH); result of Step 3.
3. the “high pass/low pass” image (HL); result of Step 5.
4. the “high pass/high pass” image (HH); result of Step 6.