Interferometric synthetic aperture radar relies on determining the digital elevation model by absolute phase, which has to be reconstructed by the process of recovering the unwrapped phase and residual phase ambiguity. The splitspectrum algorithm is one of the frequencydomain algorithms to utilize the carrier frequency in direct proportion to the interferometric phase to determine the absolute phase. However, in the process of estimating the residual integer ambiguity, the algorithm could not avoid noisy points resulting in low accuracy and efficiency. This paper proposes an improved splitspectrum algorithm. To decrease the influence of the systematic noise and decorrelation on interferometric phases with different frequencies, this paper analyses and designs a proper bandpass filter. Moreover, based on the distribution of the interferometric phase, clustering integral algorithm is proposed to determine the residual integer ambiguity. Compared with the original median algorithm, the new improvement has raised the accuracy and efficiency. Lastly, the data of ENVISAT ASAR is simulated to verify the effectiveness of the improved algorithm.