It is valuable to fuse a Green Fluorescent Protein (GFP) fluorescence image and a corresponding phase contrast image for the research of the protein function and the localization of sub-cellular structures. With the aim to eliminate the false images come from the fusion of fluorescent images and phase contrast images by the multi-scale fusion algorithm based on the ARSIS concept in remote sensing images, a variable-transparency concept is proposed.A set of functions are designed based on visual effects. Then, each pixel is assigned a multi-scaled transparency,and original images are synthesized from the results of a Stationary Wavelet Transform (SWT). Furthermore, 30 groups of images are used to estimate parameters involved in the function set and then 117 groups are assessed by means of qualitative tests. Quantified indexes related to the fused product and original images, such as the quality index Q and the high frequency correlation coefficient HPCC, are calculated for inside and outside of the fluorescent area, respectively. The experimental results demonstrate that compared with common algorithms, including the transparency method, chessboard method and the SWT substitution method, the proposed algorithm can keep the interactive user-transparency and improve the definition of fused products while reducing the effect of the dark background of fluorescent images.