I'm reading a paper about sub-pixel motion estimation optimization algorithm in HEVC; while all the proposed concepts are based on "Modeling the Error Surface" in the search range(search window)during algorithm;
Does anybody by any chance know the definition of "Error Surface" here?
And what I'm lookin for is definitely not this: Freeform surface modelling.
Thanks.
By the way, the paper's link is here.
The picture below (from the Geneva meeting in January this year) shows the
Integer samples (shaded blocks with upper-case letters) and fractional sample positions (un-shaded blocks with lower-case letters) for quarter sample luma interpolation
The following calculations are needed for the quarter-sample-interpolation:
a0,0 = ( −A−3,0 + 4 * A−2,0 − 10 * A−1,0 + 58 * A0,0 + 17 * A1,0 − 5 * A2,0 + A3,0 ) >> shift1 (8‑292)
b0,0 = ( −A−3,0 + 4 * A−2,0 − 11 * A−1,0 + 40 * A0,0 + 40 * A1,0 − 11 * A2,0 + 4 * A3,0 − A4,0 ) >> shift1 (8‑293)
c0,0 = ( A−2,0 − 5 * A−1,0 + 17 * A0,0 + 58 * A1,0 − 10 * A2,0 + 4 * A3,0 − A4,0 ) >> shift1 (8‑294)
d0,0 = ( −A0,−3 + 4 * A0,−2 − 10 * A0,−1 + 58 * A0,0 + 17 * A0,1 − 5 * A0,2 + A0,3 ) >> shift1 (8‑295)
h0,0 = ( −A0,−3 + 4 * A0,−2 − 11 * A0,−1 + 40 * A0,0 + 40 * A0,1 − 11 * A0,2 + 4 * A0,3 − A0,4 ) >> shift1 (8‑296)
n0,0 = ( A0,−2 − 5 * A0,−1 + 17 * A0,0 + 58 * A0,1 − 10 * A0,2 + 4 * A0,3 − A0,4 ) >> shift1 (8‑297)
– The samples labelled e0,0, i0,0, p0,0, f0,0, j0,0, q0,0, g0,0, k0,0, and r0,0
are derived by applying an 8-tap filter to the samples a0,i, b0,i and c0,i with
i = −3..4 in the vertical direction as follows:
e0,0 = ( −a0,−3 + 4 * a0,−2 − 10 * a0,−1 + 58 * a0,0 + 17 * a0,1 − 5 * a0,2 + a0,3 ) >> shift2 (8‑298)
i0,0 = ( −a0,−3 + 4 * a0,−2 − 11 * a0,−1 + 40 * a0,0 + 40 * a0,1 − 11 * a0,2 + 4 * a0,3 − a0,4 ) >> shift2 (8‑299)
p0,0 = ( a0,−2 − 5 * a0,−1 + 17 * a0,0 + 58 * a0,1 − 10 * a0,2 + 4 * a0,3 − a0,4 ) >> shift2 (8‑300)
f0,0 = ( −b0,−3 + 4 * b0,−2 − 10 * b0,−1 + 58 * b0,0 + 17 * b0,1 − 5 * b0,2 + b0,3 ) >> shift2 (8‑301)
j0,0 = ( −b0,−3 + 4 * b0,−2 − 11 * b0,−1 + 40 * b0,0 + 40 * b0,1 − 11 * b0,2 + 4 * b0,3 − b0,4 ) >> shift2 (8‑302)
q0,0 = ( b0,−2 − 5 * b0,−1 + 17 * b0,0 + 58 * b0,1 − 10 * b0,2 + 4 * b0,3 − b0,4 ) >> shift2 (8‑303)
g0,0 = ( −c0,−3 + 4 * c0,−2 − 10 * c0,−1 + 58 * c0,0 + 17 * c0,1 − 5 * c0,2 + c0,3 ) >> shift2 (8‑304)
k0,0 = ( −c0,−3 + 4 * c0,−2 − 11 * c0,−1 + 40 * c0,0 + 40 * c0,1 − 11 * c0,2 + 4 * c0,3 − c0,4 ) >> shift2 (8‑305)
r0,0 = ( c0,−2 − 5 * c0,−1 + 17 * c0,0 + 58 * c0,1 − 10 * c0,2 + 4 * c0,3 − c0,4 ) >> shift2 (8‑306)
Quite a mouthful as you can see...
The paper you are referring to error surface
is probably the difference between the pixel-values calculated using the method proposed in the standard and the second-order-function proposed in the paper. Hope it helps :-)