Currently, functional near-infrared spectroscopy (fNIRS) is widely used in the field of Neuroimaging. To solve the signal-noise frequency spectrum aliasing in non-linear and non-stationary fNIRS characteristic signal extraction, a new joint multi-resolution algorithm, EEMD-ICA, is proposed based on combining Independent Component Analysis with Ensemble Empirical Mode Decomposing. After functional brain imaging instrument detected the multi-channel and multi-wavelength NIR optical density signals, EEMD was performed to decompose measurement signals into multiple intrinsic mode function according to the signal frequency component. Then ICA was applied to extract the interest data from IMFs into ICs. Finally, reconstructed signals were obtained by accumulating the ICs set. EEMD-ICA was applied in de-noising Valsalva test signals which were considered as original signals and compared with Empirical Mode Decomposing and Ensemble Empirical Mode Decomposing to illustrate validity of this algorithm. It is proved that useful information loss during de-noising and invalidity of noise elimination are completely solved by EEMD-ICA. This algorithm is more optimized than other two de-noising methods in error parameters and signal-noise-ratio analysis.