Recovery of complex images from Fourier magnitude

Abstract A computational investigation of the recovery of a complex image (i.e. an image whose pixels can have arbitrary phases) from the modulus of its Fourier transform is presented. The results indicate that an acceptable level of recovery can be obtained using Fienup's iterative algorithms, provided a good, but not necessarily exact, estimate of the image support (which does not need to be of a special form) is obtainable. The problem of poor correlation between subjective restoration quality and numerically calculated error measures is addressed, suggesting a more effective combination of established iterative algorithms. Results are given for several specimen images.