Algorithm 1: Two-Level Adaptive Median Noise Removal Filter.

Input: The gray level gx,y of the pixel at the coordinates (x,y), the maximal value smax of the side of the square Us.

Output: The noise removed value of the gray level ${\overline{g}}_{x,y}$ of the pixel in the coordinates (x,y).

Step 1: Assume that LA and LB are the two levels of noise pollution.

Step 2: Consider a s-neighborhood Us. Initially, we take s = 3.

Step 3: Calculate gmed, gmin, gmax as the median, minimal, and maximal gray level of the pixels in Us.

Step 4: Level LA.

$A_{11}=g_{med}-g_{min},A_{12}=g_{med}-g_{max}.$

Step 5: if (A11 > 0  &&   A12 < 0) then go to Step 11.

Step 6: else s = s + 2 // We increase the size of s-neighborhood Us (s must be an odd number, as the pixel (x,y) must be in the center).

Step 7: end if

Step 8: if s ≤ smax then go to Step 4.

Step 9: else Output: ${\overline{g}}_{x,y}=g_{x,y}$.

Step 10: end if

Step 11: Level LB.

$B_{11}=g_{x,y}-g_{min},B_{12}=g_{x,y}-g_{max}.$

Step 12: if (B11 ≥ 0   &&   B12 ≤ 0) then Output: ${\overline{g}}_{x,y}=g_{med}$.

Step 13: else Output: ${\overline{g}}_{x,y}=g_{x,y}$.

Step 14: end if