To recover the initial signals from coupled map lattices (CML) is of great importance in signal processing. But like all chaotic systems, CML is highly sensitive to noise. It is hard for CML to recover the initial signals that have been corrupted by noise. A multi-CML system, which couples many one-dimensional CMLs paratactically together, is proposed in this paper. Through the coupled one-dimensional CML, the chaotic trajectories, which tend to be exponentially separated in noisy environment, become closer, so the noise influence is restrained. The numerical results verified the robustness and noise-restraining property of the multi-CML system. Given the proper selection of coupling parameters, even in the environment that initial signals are noise-corrupted, compared with single one-dimensional CML, the multi-CML system not only can recover the statistical property of initial conditions well, but also performs well in recovery of single initial signal with a high correlation.