This article evaluates different improvements of Laurent Itti's (Itti et al., 1998) visual attention model. Sixteen persons have participated in a qualitative evaluation protocol on a database of 48 images. Six different methods were evaluated, including a random fixations generation model. A real time conspicuity maps generation algorithm is also described. Evaluations show that this algorithm allows fast maps generation while improving saliency maps accuracy. The results of this study reveal that preys / predators systems can help modeling visual attention. The relatively good performances of our centrally biased random model also show the importance of the central preference in attentional models.