“…It is also worth noting that in the use of SSIM in video coding, a number of variations have been employed in the computation of SSIM to better fit the computational structures of video coding algorithms. For example, as opposed to the 2D Gaussian sliding window employed in the original SSIM computation [6], different sizes of square windows have been used to calculate block SSIM, including 16×16 sliding windows in [12], [10], [11], [15], 8×8 sliding windows in [13], [14] and 8×8 non-overlapping windows in [16], [19], [20], [21]. In [12], [10], [11], [15], [16], [19], SSIM of a block is obtained by averaging the SSIM values of its 4×4 non-overlapping child blocks, in [13], [14], block SSIM is calculated in the full current block directly, while in [20], [21], 4×4 sliding windows are used for SSIM calculation in an extended block.…”