In this paper, we propose a novel method for image separation combining wavelet transform and ICA with reference. For the sake of simplicity, we do not consider the influence of noise and named this new algorithm as Wavelet-ICA-R. First, three mixed images are obtained by mixing three original images together, utilized a randomly generated mixing matrix. Then, transform the reference signal and the mixed images from time domain to wavelet domain respectively. Finally, in the wavelet domain, exploit 1/16 of the high frequency reference information to extract the original images. Simulation results indicate that this algorithm shortens the calculated time by 60% on average than ICA with reference while exhibits a superior performance.