An approach for the removal of impulse noise from highly corrupted image sequence is proposed to achieve optimal image quality as well as fast computation time. Using 3-D image data which is spatial and also temporal pixel, the PSNR of image sequence is increased and the number of computation is decreased. This proposed algorithm is compared with the widely used impulse noise removal algorithms in aspect to the PSNR and also the number of computation. We propose the detection method of the corrupted pixel by impulse noise. Using the proposed method, we can preserve the pixel which is not corrupted, so the number of computation is dramatically decreased. This method has a good result in low ratio of impulse noise, and especially, has the best quality in no corrupted sequence. In this algorithm, the frames of sequence are divided into the highly corrupted frames and the frames which are corrupted by impulse noise under 12%. For the highly corrupted image frames, we make use of the linear operation based on FIR - median hybrid operation.