The light waves diffracted from objects contain the amplitude and phase information. The light intensity information can be recorded easily by current cameras, however, the amplitude information has the worse ability to resist noise and is easily to lose in the process of beam propagation, so the recovery of phase information is of great significance. Through the transport of intensity equation and the angular spectrum iteration algorithm based on Gerchberg-Saxton (GS), the retrieval of phase information in two-dimensional images was studied. Firstly, two images were used to simulate the amplitude and phase information of the light wave at the position of the imaging plane. Then the complex amplitude distribution on two defocused planes in the imaging optical path were calculated taking advantage of the angular spectrum transfer function. The light intensities of three different parallel planes can only get rough phase by the transport of intensity equation due to the multiple approximations. The rough result were then used as the initial phase values in the GS angular spectrum iterative algorithm to speed up the convergence of iteration and finally obtain the more accurate phase distribution. The root mean square of the phase error in iteration is as low as 10^-8 in the Matlab simulation. By applying the method to three different colors and then combining them together, the phase and intensity of color images on any planes can be obtained by only detecting the intensities on three different planes.