In this study, we develop a BM3D-based algorithm to improve the performance of color computational ghost imaging. Specifically, we respectively project the speckle patterns of red, green and blue onto the object. Then, we adopt the single-pixel photodetector to record the total light intensity reflected by the target object. Finally, we calculate the correlation between the light intensity sequence and the corresponding speckle patterns based on BM3D algorithm. Our simulation results show that the developed scheme can achieve high-quality reconstruction image even at low sampling ratio.