Non-linear Techniques For Image Restoration

Abstract

The ‘absolute’ method (AM) of superresolution has been applied to two-dimensional, computer-synthesised images depicting a realistic military scenario. Preprocessing, by median filtering, of a cloudy sky image containing an indeterminable number of single-pixel point sources in a 256 x256 image firstly isolated the point sources from the background, such that the point sources were based on a zero background level. The point sources may contain either a single point, or two closely spaced point sources, which are unresolved in the original image due to the diffraction limit of the simulated imaging process. Wiener deconvolution was used to restore the image within the passband, resulting in an extremely high level of high frequency oscillation, shrouding the restored points of interest. Application of five iterations of the Ah4 simultaneously suppressed the noise and enhanced the amplitude of the correctly resolved point sources. The method was shown to successfully nestore single pixel point sources in an image, even if the point sources were located with as few as two pixels separating them. Six pixel separation of the points is about ithe limit that could be resolved with the naked eye.