Hardware-accelerated objective function for medical image registration
W. Withayachummankul, B. Laksanapanai, Chuchart Pintavirooj · 2004
Generally, image registration using genetic algorithm is a time-consuming process since the algorithm needs to evaluate the objective function several hundred times depending on the vastness of search space. The situation appears worse if the registration is intensity-based due to the interpolation loops prior to each objective function. However, with the availability of modern graphic processing unit (GPU) one can lake an advantage from interpolating division normally used to interpolate the texels for 2D-texture mapping. Issued via OpenGL API, the image interpolation requires half time less than it ever does without losing of accuracy.