Influences of infrared image complexity on the target detection performance

To establish quantitative, accurate, and strict function between infrared image complexity metrics and target detection performance, and to carry out strict mathematical proofs is one of the problems urgently to be resolved in the infrared target detection field at present. Zero-mean normalized cross-correlation algorithm was used to detect target, partial least-squares was used to establish functional models between three infrared image complexity metrics and two target detection performance indexes simultaneously, which included all the significant image metrics for target detection performance, multicollinearity problem among image complexity metrics had been solved effectively. Cross validity criteria, adjusted multiple correlation coefficient, and F-test were adopted to test the significance of the regression equation. Spearman rank correlation coefficient and mean relative error were adopted to test the prediction performance of the regression equation. The results show that the established regression equations are highly significant, their goodness of fitting is better, their prediction performance meet certain requirements, and the ways to further improve the performance of regression models are analyzed.