People with autism spectrum disorder (ASD) tend to detect local patterns of visual stimuli more quickly than global patterns, which is opposite to the behavior of typically developing people. We hypothesized that the imbalance between excitation and inhibition neurons in the visual cortex causes the local processing bias observed in ASD. Stronger inhibitory connections could diminish the neural activities and thus prevent global feature integration, whereas properly balanced connections would enable the cortex to detect features of any size. We verified our hypothesis by employing a computational neural network called a neocognitron. Our experimental results demonstrated that the network with stronger inhibitory connections exhibited a local processing bias, whereas the network with properly adjusted connections showed a moderate global bias. Moreover, the networks with extremely strong or weak inhibitions revealed no perception bias. These results suggest that an excitation/inhibition imbalance causes multiple types of atypical perception in ASD.