A methodology forprocessing images of diesel sprays under different experimental situations is presented. The new approach has been developed for cases where the background does not follow a Gaussian distribution but a positive bias appears. In such cases, the lognormal and the gamma probability density functions have been considered for the background digital level distributions. Two different algorithms have been compared with the standard log- likelihood ratio test (LRT): a threshold defined from the cumulative probability density function of the background shows a sensitive improvement, but the best results are obtained with modified versions of the LRT algorithm adapted to non-Gaussian cases.
© 2007 Optical Society of America
Original Manuscript: May 30, 2006
Revised Manuscript: October 20, 2006
Manuscript Accepted: October 23, 2006
Published: February 2, 2007
José V. Pastor, Jean Arrègle, José M. García, and L. Daniel Zapata, "Segmentation of diesel spray images with log-likelihood ratio test algorithm for non-Gaussian distributions," Appl. Opt. 46, 888-899 (2007)