A novel adaptive scene-based nonuniformity correction method for fixed-pattern noise removal was put forward. It is based on the multiframe registration and estimation of global translation between several adjacent frames. The resulting mean square error function was optimized making use of the least mean square algorithm. Combining the local variance of an error function with the correlation peak value, the convergence speed can be controlled adaptively. The proposed method takes advantage of the correlation of adjacent frames sufficiently, thus it can provide enhanced results for diverse simulated and real infrared image sequences with nonuniformity. The experimental results show that the accurate estimations of the nonuniformity parameters of each detector in a focal plane array speed up the convergence, meanwhile, retain few ghosting artifacts.