Chapter 3
Blind Deconvolution Methods: A Review
In this chapter we look at the different solution methods that have been proposed for
blind deconvolution. Though the methods could be classified in many ways, in order
to depict the evolution of the solution methods, we describe the different techniques
under a broad classification of earlier and recent techniques, which is further split
into deterministic and statistical techniques.
3.1 Earlier Approaches
Spectrum of image with motion blur caused by translational motion has spectral
zeros and earlier approaches for estimating the PSF relied on these spectral zeros
to estimate the blur and hence were applicable only for a limited type of blurs. We
describe classical transform based methods in this section.
One of the earliest works reported in 1975 [153] uses homomorphic signal
processing for deconvolution. Here in noise free case the reconstruction filter is the
inverse filter and it is shown that in the noisy case it becomes the geometric mean of
the inverse filter and the Wiener filter. To estimate the inverse filter multiple images
blurred by the same PSF is required. Since this is not available the estimation is
done by splitting the observed image into smaller images with size much larger than
the PSF size to avoid edge effects. The log spectrum of these smaller images are
used to estimate the PSF [153]. It is important that the phase information is known.
Hence this method works only for simple blurs. Besides, this method works only if
an identical recording system with a flat frequency response is available. Since this
method is too limited to be of any practical use, more sophisticated methods were
developed. The classical iterative methods are described next.
One of the first iterative methods for blind deconvolution [83] uses the idea
of zero sheets [84]. It is shown in [83] that for a noise free case the individual
components of a composite image can be recovered without using any filtering
© Springer International Publishing Switzerland 2014
S. Chaudhuri et al., Blind Image Deconvolution: Methods and Convergence,
DOI 10.1007/978-3-319-10485-0__3
37