Unwrapping Hartmann-Shack images of off-axis aberration using artificial centroid injection method

As the degree of aberration and noise increases, particularly for off-axis aberration, wavefronts of Hartmann-Shack images become so distorted that special care needs to be considered in order to successfully and gracefully unwrap the images. This paper proposes an alternative algorithmic approach called the artificial centroid injection method. Initial centroid extraction is done using Laplacian of Gaussian (LoG) and dynamic thresholding. Outlier centroids are filtered using ensemble of weak classifiers boosted with Adaboost algorithm. Observing the nature of the vertical and horizontal centroid sequences using Kalman Filter, multiple General Regression Neural Networks (GRNN) are then trained to approximate centroid sequences. Artificial centroids are generated by taking the intersection points of approximated vertical and horizontal GRNN sequences that occurs inside an elliptical Region of Interest optimized with Regrouping Particle Swarm (RegPSO). These artificial centroids are injected to the intial centroid vector to predictively recover missing and previously unrecognized spots. Wavefront algorithm is then applied to correspond detected centroids to their appropriate lenslet centers. This algorithm has successfully unwrapped 29 different off-axis aberration HS images, −50° Temporal plane to +50° Nasal plane up to zero pixel prediction error, with no false correlations in any of the tested images.