Study presents a novel algorithm to improve the frequency bandwidth of the seismic data through removal of non-stationary wavelet in cepstrum domain. Algorithm incorporates useful properties of both homomorphic deconvolution and spectral decomposition by this way it estimates and filters the wavelet at each translation of spectral decomposing Gaussian window which improve the resolution and tolerates the non-stationary properties of seismic wavelet. This study describes the comprehensive mathematical formulation of the algorithm and its testing on 1D, 2D synthetic and real seismic section to confirm its applicability. Comparative testing shows that the algorithm effectively removed the smearing effect of wavelet which led to broader frequency spectrum and improved seismic resolution. © 2014 Academic Journals Inc.