Raman spectrum is the characteristic spectrum for material structure and composition research as its frequency, intensity and polarization can characterize the peculiar properties of the scatterer, and thus it has been widely applied in many fields. However, the baseline drifting often occurs in the Raman spectrum and adversely affects its quantitative analysis. In order to eliminate the baseline drifting, a correction algorithm is presented, which combines the peak detection method based on derivative spectrum and the Whittaker smoother for baseline estimation. The spectral peak region is detected and identified utilizing the second derivatives. Then the Whittaker smoother calculates the curve fitting of the non- peak region combining the identification information, and interpolates the peak region baseline smoothly at the same time. The whole baseline estimation of the spectrum is obtained. The algorithm has been applied to simulated and actual Raman signals to correct the baseline drifting. The results show that the algorithm realizes spectral denoising and baseline estimation simultaneously. The improved analytical results of principal component analysis also verify its effect.