Saliency detection performances on real images under different SNRs (a) 3 examples of testing images. Up: the banana in terrain background at the SNR 1:3, mid: the pear in brick background at the SNR 1:2.5, down:the red onion in water background at SNR of 1:3.5; (b) saliency maps of the attention model with invariant coding; (c) without invariant coding; (d) saliency maps of the Saliency Tool. In a saliency map, the more salient part is indicated by the brighter and whiter region; (e) plot of the correct detection rates versus SNRs. Larger RF size improves the performance of the model with invariant coding, which coincides with findings in neuroscience.
Qi et al. BMC Neuroscience 2012 13:145 doi:10.1186/1471-2202-13-145