Efficient Parallel Implementation of Active Appearance Model Fitting Algorithm on GPU

Algorithm 1

CPU-based AAM fitting algorithm.

// Max_iter: the maximum number of iterations.

// p: the estimate of model parameters.

// Δp: the variance of model parameters.

// r: the residual texture.

// E and E′: the error between the model and image.

// R: the gradient matrix.

// Thr: the threshold of the variation of error.

(1)   K []←{1, 0.5, 0.25, …};

(2)   𝑓𝑙𝑎𝑔←true; 𝑖𝑡𝑒𝑟←1;

(3)  estimate the model parameters, 𝑝;

(4)  compute initial 𝐸;

(5)   while  flag && iter <= Max_iter

(6)      compute Δp: Δp ←  − 𝑅𝑟(𝑝);

(7)       for each k in  K

(8)       𝑝←𝑝 + kΔp;

(9)       compute new E′;

(10)         if  𝐸^′<𝐸

(11)          stop for loop;

(12)         end  if

(13)       end  for

(14)       if  𝐸^′−𝐸 < Thr

(15)        flag ← false;

(16)       else

(17)        𝐸←𝐸^′;

(18)       end  if

(19)      𝑖𝑡𝑒𝑟←𝑖𝑡𝑒𝑟+1;

(20) end  while