A time series of experimental optical chaos signal with dynamic equation unknown and low SNR is obtained. The wavelet multi-resolution decomposition algorithm is applied here to reduce the noise mixed in the experimental optical chaos signal. The performance of the algorithm is verified by Lorenz chaos signal mixed with noise, which shows that the SNR is increased by 10dB or so. Some parameters of the optical chaos attributes are calculated before and after noise-reduction. It shows that the noise-reduction algorithm can improve the precision of the Lyapunov exponent calculated with small data method, and a completely opposite wrong result can be avoided by the noise-reduction process when computing the minimum embedding dimension with Cao method. The small data sets method is improved by Cao method (minimum embedding dimension) and mutual information method (delay time). As the result is shown, the error of the largest Lyapunov exponent is reduced by nearly 30%, and the largest Lyapunov exponent of the optical chaos signal is 0.3896 obtained with this method.