Fast Implementation of Spatial Averaging

We present a new fast spatial averaging (FSA) technique that efficiently implements operations for spatial averaging or mean filtering. To perform spatial averaging of an M x N image with an averaging filter of size m x n, the basic method requires (mn 1) MN additions and MN divisions, while recent work using the column-addition and store-and-fetch techniques reduces the number of additions to approximately (m + 1)MN and eliminates the need for division. Our proposed method further reduces the number of redundant additions and requires only 4 MN additions and no division. Our FSA algorithm is very promising since the major computations it requires do not depend on the size of the averaging filter. Experimental results on various image sizes using different filter sizes demonstrate that our method is significantly faster than other spatial averaging algorithms.