A spectral correction algorithm for correcting dense object-induced spectral artifacts is described. In one embodiment, a calibration object, representative of typical head scanning conditions is scanned and the data reconstructed to provide an image. A water or water-equivalent cylinder of about the same diameter also is scanned and reconstructed, on the same display field of view (DFOV). These two images are designated respectively by BWEQ and WEQ. The ratio of images BWEQ and WEQ is then evaluated, and a region of interest extracted by multiplying the ratio by a function II(r), to obtain a calibration pattern CP. The calibration pattern is then averaged in azimuth to obtain a calibration vector. This calibration vector is fitted with low--order polynomial, and then divided by the fitting polynomial, to take out from the vector the low frequency components that, for instance, would be introduced on an "ideal" scanner. By subtracting 1.0 from the ratio, and multiplying by a CT number scale factor (ctscale) and an apodizing window Aw(r), a calibration error vector CEV is obtained that is representative of the circularly symmetric image error introduced by the non-corrected bone-induced artifact. The corresponding error calibration vector can be expanded into a circularly symmetric image error pattern I[CEV(r)], and subtracted from the calibration image, to provide a substantially artifact free image. The method can be extended to extract and correlate error vectors on an image segment basis such that the resulting error image pattern is not circularly symmetric.