Illustrative examples of DCT coefficients of residual noise, obtained by denoising image. The same two DCT subbands (a, b), as in Figure 2 are extracted the residual noise of lena image. On those examples, the assumption of i.i.d. distribution seems to be more realistic. It is noted that three denoising filters are respectively designed based on wavelet, block-matching and 3D (BM3D) and dictionary learning (K-SVD) algorithms.