From the data with low signal-to-noise ratio (SNR), such as intrinsic signal of optical imaging (OIS) data, the extracted response signal by blind source separation (BSS) method is usually distorted seriously. To overcome this shortcoming and obtain high-quality signal sources, we propose a procedure to increase the SNR in signal source separation. Our basic idea is that, finding out and enhancing the pixels which probably contain the signal of our interest. Because these locations accommodate intense interesting signals, intensifying the signals located in them is promising for increasing the SNR. The initial low-quality signal of interest and their approximate locations can be obtained by the conventional BSS method. The approximate locations are used to increase the SNR of the raw data. From the SNR-increased data, one can obtain more accurate interesting signals and their locations by a conventional BSS method. By feeding back these more accurate locations, SNR can be increased again. Result on simulated data illuminates that such iterative procedure greatly increased the SNR, and comparison to other algorithm shows that our method is efficient in recovering the underlying signal sources with high qualities.