An infrared noise reduction algorithm based on anisotropic partial differential equation is proposed. By combining morphological operations with infrared imaging local features, a new anisotropic coefficient is established. This coefficient gets gradient operator by morphological dilating and eroding, improves the sensibility of Perona-Malik (P-M) gradient, strengthens diffusion in homogeneous area and weakens diffusion in heterogeneous area. The algorithm has run on the EVM-DM642 hardware platform in real time, and experiments indicate the noise is smoothed out effectively, and the image edge is also preserved better.