Adaptive windowing in motion detector for deinterlacer

1. A method of determining a motion value for a missing pixel in an interlaced video field using an adaptive window, the method comprising:

computing a first mean absolute difference (MAD) value for the pixel based on a 1×5 window;

computing a second MAD value for the pixel using a 3×5 window; and

selectively blending the first and second MAD values to form a resultant motion value;

wherein selectively blending includes computing a blend factor blend according to:

  topbot_diff1 = abs((next_top + prev_top)/2 − (next_mid +

prev_mid)/2));

  topbot_diff2 = abs((next_bot + prev_bot)/2 − (next_mid +

prev_mid)/2));

  topbot_diff = max([topbot_diff1 topbot_diff2]) −

topbot_thresh;

blend = topbot_diff + blend_res/2;

if (blend<0)

  blend = 0;

elseif (blend>blend_res)

  blend = blend_res;

end;

wherein next_top, next_mid and next_bot correspond to pixels in a next field, prev_top, prev_mid and prev_bot correspond to pixels in a previous field, the previous field having a same parity as the next field and an opposite parity to the interlaced video field of the missing pixel;

and wherein topbot_thresh and blend_res are programmable registers.

2. The method of claim 1 , wherein selectively blending further includes:

forming the resultant motion value frame 02 using the blending factor blend according to

  frame02 = round(((sad02_win1x5)*3/16 * blend) + (sad02_win3x5

/16 * (blend_res-blend)))/blend_res;

wherein sad 02 _win1×5 is a sum of absolute difference corresponding to the first MAD value in the 1×5 window and sad 02 _win3×5 is a sum of absolute difference corresponding to the second MAD value in the 3×5 window.

3. The method of claim 2 , further comprising:

normalizing each of the first and second MAD values mad_curr as follows:

  mad_curr_cored = round((mad_curr-mad_core_thresh)*

mad_core_normalize);

wherein mad_core_thresh and mad_core_normalize are programmable registers, and mad_curr_cored is made to saturate at a value of 16.