In this paper, pixel based matching methods, and the fact that registering accuracy of phase values, for their direct function in DEM extraction, has more impact on DEM results are discussed. Then the authors analyzed the disadvantages of traditional phase average gradient function method, which can be concluded as human interference
required and its sensitivity to noises inducing low accuracy. To get avoid of those shortcomings, a modified phase average gradient function method was proposed in which the theory of pyramid matching was also brought in to make sure of reliability and lessen the amount of computation. Experiments are carried out and results, in the form of coherence maps, of the new method were compared with that of accuracy testified largest correlation coefficient method
in existed software. And the conclusion that it can gain better results was drawn. At last, further developments that can be made in the future were discussed.