An accurate Harmony Search based algorithm for the blind deconvolution of binary images

Abstract

In this paper we introduce the Largest Error First Harmony Search (LEFHS) algorithm for the deconvolution, or deblurring, of binary images. The algorithm requires no prior information about the source of the blur which is commonly modelled as a point spread function (PSF). LEFHS is able to fully recover the original binary image even when the PSF is unknown making it a true blind deconvolution algorithm. LEFHS addresses a major limitation that is present in another blind deconvolution algorithm called CHS. CHS is also able to fully recover the original binary image but only when the blur source is restricted to a binary PSF. LEFHS does not have this restriction and successfully recovers images blurred by arbitrary PSFs. LEFHS addresses a significant restriction in CHS as non-binary PSFs are the norm in real images. We compare our results with CHS and other state of the art binary deconvolution algorithms and find that LEFHS is faster than CHS even when using more complex non-binary PSFs as well as being more accurate than other binary deconvolution algorithms.