To improve the real-time performance of frame-sweep infrared remote-sensing image mosaic，LSH(locality sensitive hash) algorithm was used for fast image registration，and the LSH performances based on Hamming distance、Euclidean distance and cosine distance were compared.The results show LSH algorithm based on hamming distance has better real-time performance and better precision in infrared image registration.By implementing and improving the LSH algorithm based on Hamming distance，and using infrared remote-sensing images to validate the performance of image real-time registration，the real-time infrared image mosaic system based on Hamming-LSH is realized.