Baseline correction is an important step of spectral retrieving procedure for Fourier transform imaging spectrometer data processing, because preprocessing steps of spectral retrieving procedure, such as interference curve apodization and phase correction, need to be implemented after baseline correction is completed. A selfadaptive differential filtering method for baseline correction is presented. This method uses an iterative algorithm for dynamically adjusting its weighted mean filter window. The simulation results demonstrate that the filtering method can filter out the direct current trend more thoroughly. Analysis of the ultraviolet spectral data of latent fingerprint residues obtained by instrument in laboratory is made. The results indicate that within the working spectrum range of the instrument, the spectra obtained by the proposed baseline correction method are substantially consistent with the actual spectra. Since the method does not require a pre-selected mean filter window, it is selfadaptive. Moreover, the self-adaptive differential filtering method based on mean filter algorithm is simple and has high efficient iteration.