Infrared and visible image registration has become a big challenge in online monitoring system for high voltage electric equipment. To address this problem, an efficient registration method based on modified features from accelerated segment test (FAST) in non-subsample Contourlet transform (NSCT) domain is proposed. Firstly, the two images are preprocessed by gray-level equalization. After that, NSCT decomposition is applied to obtain low-frequency subband images. And corner points are detected from the low-frequency images as interest points by FAST. Their descriptors are calculated by partially intensity invariant feature descriptor (PIIFD) and used to make rough matching through best bin distance ratio. Finally, random sample consensus (RANSAC) is further utilized to refine the matched interest point pairs and affine transformation parameters between infrared and visible images are determined. Subjective and objective evaluation on experimental results shows that the proposed method achieves significant improvement in registration accuracy, computation speed and interference robustness with its practical value.