Saliency detection in visual search where stimuli differ in orientations. (a) a testing image without orientation noises; (b) an example of testing images where orientation noises are in [−30°, 30°] with uniform probability. The target marked by red circles; (c) comparison of different models namely the model with invariant features,the model without invariant features, a fully connected network and a randomly connected network, on 20 testing images such as that in (b) when layer 1 RF size varies; (d) comparison when layer 2 RF size varies; (e) comparison when the coefficient αwhich adjusts the suppression strength varies; (f) comparison when excitation bandwidth σ
varies. See the text for details.