Abstract
Discriminating between object categories is a critical function of primate and bird visual systems. In primates, neurons at each stage of the ventral visual stream pool the responses of neurons at lower stages of processing and represent increasingly complex information. Birds display a radically different neural architecture to primates and no structures correspond directly to the ventral stream regions. Despite these superficial differences, the avian visual system is composed of a cortex-like circuitry that may mimic the computational principles of the mammalian brain. In addition to displaying an efficient neuronal organisation comparable with visual cortex, avian vision is on par with that of primates at the behavioural level. The aim of the studies in the current thesis was to determine whether a similar hierarchical organisation that operates across the ventral stream in monkeys supports face and object categorisation in the avian visual system.
We performed electrophysiological recordings from visual nuclei of the pigeon visual network to compare the neural coding mechanisms with the primate visual cortex. After determining that visual categorisation tasks are not sufficient for assessing neurons purely sensory properties, we trained pigeons to view and withhold responses to images presented in the frontal visual field without requiring the birds to behaviourally categorise them. We found evidence of a hierarchical organisation between the entopallium (ENTO) and mesopallium ventrolaterale (MVL) that recodes visual information for more complex shape identification in the pigeon brain. In contrast with higher stages of the primate ventral stream, neurons responded selectively to images’ local features as well as faces at the level of MVL. While some neurons responded best to faces over other stimuli, the degree of selectivity for faces was not equivalent to primate face-selective neurons. Further investigation of neurons responses to progressively scrambled images of objects confirmed that MVL represents both local and global image features. In summary, the studies of the present thesis demonstrate that the avian visual system recodes information in a comparable series of processing stages to the mammalian visual cortex, but suggests that the neural mechanisms supporting object recognition differ substantially from the mammalian brain.