A new method for evaluating image-recovery algorithms is presented, which is based upon the numerical assessment of the performance of a specified visual task. In this method a Monte Carlo technique is used to simulate the full imaging sequence including the generation of scenes appropriate to the desired application, subsequent data taking, image recovery, and analysis of the final images. The evaluation of the algorithm is made on the basis of a task performance index, which measures how well the task is performed. The usefulness of this method is demonstrated by a study of the algebraic reconstruction technique (ART) for the recovery of an image from its projections. Comparison between the detection results achieved with images reconstructed with unconstrained and constrained ART shows that the nonnegativity constraint is generally useful in improving detectability, especially when the projection data consist of a limited number of noiseless projections.
Keywords: evaluation of image-processing algorithms, task performance, algebraic reconstruction technique (ART), nonnegativity constraint, detectability, detectability index, receiver operating characteristic (ROC), Monte Carlo technique
Get full paper (pdf, 1229 KB)
Return to publication list
Send e-mail to author at kmh@hansonhub.com