Multispectral IR imaging techniques are frequently deployed in maritime operations, for instance to detect floating mines or to find small dinghies and swimmers during search and rescue operations. However, maritime backgrounds usually contain a large amount of clutter that severely hampers the detection of dim point targets. Here we present a simple algorithm that deploys the correlation between target signatures in two different (3-5 and 8-12 mum) IR frequency bands to reduce the amount of clutter. First, both individual IR bands are filtered with a morphological opening top-hat transform to extract small details. Second, the resulting detail images are thresholded to produce binary detail images, representing potential target areas. Third, a fused detail image is obtained by taking the intersection (logical AND) of both binary IR detail images. Details that appear in both IR bands remain in this fused detail image, whereas a large fraction of uncorrelated noise details is filtered out. Remaining noise details can be removed by taking into account the temporal characteristics of the target signatures and by using a priori knowledge of structure of the scene and the size of potential targets. The method is tested on two image sequences showing a maritime scene with three kayaks approaching from far away. The scenario was registered in the 3-5 mum and 8-12 mum IR frequency bands, and in the visual range. The results show that the proposed multispectral processing technique has the potential to improve the detection of dim point targets in cluttered maritime backgrounds.