“…I A (x, y; β i ) = |M(x, y) * H (x, y; β i )| 2 = [M(x, y) * H (x, y; β i )][M(x, y) * H (x, y; β i )] * = [M(x, y) * H (x, y; β i )][M(x, y) * H * (x, y; β i )] (A.3) since M(x, y) is real.Therefore, the partial derivative of the aerial image with respect to the variable θ is∂ I A (x, y; β i ) ∂θ( p, q) = [M(x, y) * H (x, y; β i )][M(x, y) * H * (x, y; β i )] ∂ M( p, q) × ∂ M( p, q) ∂θ( p, q) , = {[M(x, y) * H * (x, y; β i )]H (x − p, y − q; β i ) + [M(x, y) * H (x, y; β i )]H * (x − p, y − q; β i )} × − sin θ( p, q) 2 , (A.4) because M( p, q) = (1 + cos θ( p, q))/2.Since Î (x, y), I (x, y; β i ), I A (x, y; β i ), H (x, y; β i ), and M(x, y) are matrices of many variables, we use their short forms Î , I , I A , H and M for convenience in the following derivations. Given the cost function derivations (equations (A.1)-(A.4)), the gradient∇ F is therefore ∂ F ∂θ( p, q) − Î )sig(I A )[1 − sig(I A )] ∂ I A ∂θ( p, q) I − Î )sig(I A )[1 − sig(I A )][(M * H * ) × H (x − p, y − q; β i ) + (M * H )H * (x − p, y − q; β i )] sin θ( p, q) 2 = −α i η i {H (β i ) * [(I − Î ) I (1 − I ) (M * H * (β i ))] + H * (β i ) * [(I − Î ) I (1 − I ) (M * H (β i ))]} sin θ, (A.6)which is the expression in equation(13).…”