Digital Signal Processing Reference
In-Depth Information
capturing better contours. The similar comparison results are obtained when recon-
structing other images. We truncates a part of performance comparison values ob-
tained from reconstructed images using 10 percent of transform domain coefficients
and show them in Table 1.. The PSNR and time values in Tab. 1. indicates the trend
that on the whole, WBCT is more excellent than the other two transforms.
Fig. 5.
The PSNR-coefficient curves for the
reconstructed results
Fig. 6.
The time-subrate curves for the recon-
structed results
Table 1.
Performance comparison (coefficient ratio=10%)
Item
Image
PSNR(dB)
Time(sec)
WBCT
Contourlet
DWT
WBCT
Contourlet
DWT
barbara
22.69
22.46
22.33
34.10
37.53
74.33
baboon
21.72
21.63
21.56
25.58
38.93
89.23
boat
25.01
25.06
24.80
32.49
35.52
59.65
goldhill
26.77
23.48
26.83
56.84
42.99
60.59
lena
28.05
27.46
27.59
46.03
52.20
60.28
peppers
27.03
26.25
27.20
108.58
106.12
130.81
5
Conclusion
This paper has proposed a scheme of compressed sensing employing the sparse repre-
sentation of WBCT. Sparse representation is one of the key elements of CS, so we
should select proper basis to preserve details and contours of images. Having the
advantage of non-redundancy and being capable of capturing finer contours, WBCT
is relatively a perfect sparse transform. In our experiments we have adopted the
general paradigm of BCS coupled with an IHT reconstruction using WBCT sparse
basis promoting not only sparsity but also smoothness of the reconstruction.
