Fig. 4.

2D linear fit method (A, red closed circles) was more precise than the central difference (B, black squares) or Sobel filter (C, blue open circles). Dashed cross-lines are the set value for the gradient (Gx = 0.43 ms/pixel, Gy = 0.25 ms/pixel), and the dashed circles represent mean ± 2 × S.D. (n = 100, 5 × 5 window size, SNR = 4) of each method. Each method became more precise with larger window sizes (D), and solid lines in (D) were trend lines using power-law functions. The linear fit method had the smallest standard deviation in all conditions.