Compared with the traditional computed tomography (CT), the spectral CT can obtain projection images of the object in different energy spectrum channels with a single scan, which is helpful to improve the contrast-to-noise ratio and distinguish the materials. Spectral CT based on photon counting detector is a hot research topic in recent years. As the energy spectrum channel narrows, the noise increases in each energy spectrum channel. In order to reduce the noise in the channels effectively, the Split-Bregman algorithm based on the total variation minimization is used for spectral CT image reconstruction. The spectral range is divided into different channels according to the prior information of the reconstructed model. The reconstructions are conducted for the projection data with noise and sparse angle based on the Split-Bregman algorithm. The simulation results show that the spectral CT image reconstruction based on the Split-Bregman algorithm can reduce the influence of the noise in spectral channels effectively, and satisfying the requirement of substance distinguishing.