The recovered image in ghost imaging (GI) contains an error term when the number of measurements M is limited. By iteratively calculating the high-order error term, the iterative ghost imaging (IGI) approach reconstructs a better image, compared to one recovered using a traditional GI approach, without adding complexity. We first propose an experimental scheme, for which IGI can be realized, namely the narrowed point spread function and exponentially increased signal-to-noise ratio (SNR) are realized. The exponentially increasing SNR when implementing IGI results from the replacement of M with M(k). Thus, a perfect recovery of the unknown object is demonstrated with M slightly bigger than the number of speckles in a typical light field. Based on our theoretical framework from the angle of high-order correlation R(k), the two critical behaviors of the iterative coefficients α and the measurements M are derived and well explained.