The Local Standard Deviation (LSD) images play an important role in the spatial coherence test with satellite imagery. The traditional sliding window approach, however, is very slow in computing LSD owing to its longer iteration times during operation. Especially, in the case of large imagery and small size of the sliding window, the operation is remarkably time consuming. A fast algorithm for calculating LSD is presented based on the idea of matrix operation. The first step of the algorithm is to shift the raw image matrix in different directions to generate a series of new image matrices. Then, LSD is calculated by using mathematic operation on the shifted image matrices generated in firs step. Finally, the NOAA 16/AVHRR images measured on January 1, 2005 are used as an example to calculate the LSD of Channel 4 brightness temperature by using the fast algorithm presented and the sliding window approach, respectively. The results show that the compute efficiency of the fast algorithm is evidently higher than that of the sliding window approach.