We have studied the capability of the AWS model of
predicting human fixations. Here we show the results obtained on two
open-access datasets, from Bruce
& Tsotsos (2006) and from Kootstra
et al. (2008). Through a ROC analysis (Tatler
et al. 2005), we have computed the average of the AUC and the corresponding standard error, using the same setup that has been
employed by Seo & Milanfar (2009) and Zhang
et al (2008), for a fair comparison with reported results for other models.
* Results published by Zhang et al. (JoV 2008) (using the same procedure).
**Results published by Seo & Milanfar (JoV 2009) (using the same procedure.)
- For the results of Itti et al. (PAMI 1998) we have used the matlab code of the original version provided by J. Harel, instead of the STB (with a different setup) used in most papers.
-Standard errors for both of the datasets are in the range 0.0007-0.0008 for all models. For the groups of Kootstra et al. dataset, standard errors are in the range 0.0010-0.0018.