Infrared focal plane array IRFPA bad pixel correction in complex applications and rich details conditions is a common problem. Therefore, a scene-based bad pixel dynamic correction (SBBPDC) technique was investigated. Based on the adaptive median filtering bad pixel detection algorithm, using the key frame technique according to the scene change, the interframe statistics are applied to judging bad pixels for higher detection accuracy. Using the improved Nagao filtering bad pixel compensation algorithm with some stronger directivity filtering sub-windows which have the same pixel amount, the image boundary pixels didn't reappear as flicking dot during the quick scene changes, and the rich scene details were preserved. Three new objective evaluation parameters were proposed: detection missed factor, detection false factor and bad pixel reappearing factor to evaluate the validity and over-correction to details of algorithm. The correction experiment of actual IR images from a cooled medium wave IRFPA proved SBBPDC algorithm effective to solve the details missing and bad pixels reappearing problems existing in traditional bad pixel correction algorithms.