An improved infrared and visible image fusion method is proposed to solve problems that comprise low definition, fuzzy contour and poor adaptability for traditional image fusion methods. By mean of NSCT scheme for decomposing infrared and visible images, the sub-band coefficients of bandpass and low-frequency are obtained, respectively. MSF-PCNN is used to obtain high-frequency fusion coefficient, and provide the performance improvement of PCNN, namely NL-PCNN, which is to gain low-frequency fusion coefficient. According to two different data sources of low-frequency domain, the improved filters, such as the sharp-weighted filter and mean-weighted filter, have the ability of separating the frequency gradient to extract the profile. Experimental results show that the proposed fusion algorithm is effective. It offers advantages over the traditional image fusion algorithm for obtaining image contour information and enhancing the fusion image clarity, which has high adaptive ability.