Visual object recognition: Timing and occulsion-related effects

Modelers of the visual system are often guided by experimental results which constrain the visual system's functional architecture.{09}In the experiments reported here, we investigate two issues constraining the functional architecture of the visual system---the speed of object recognition and the performance of object recognition on partially visible objects. Recent EEG results in humans have suggested that the earliest differences between target and nontarget categories of images are stimulus-locked and arise early enough to impose severe constraints on the use of time-consuming processes such as feedback and rate-based spiking codes in the visual processing leading to object recognition. We investigate these early differences and find that they do not rely on categorization of the stimulus, but appear to be the result of low-level processing of differential statistical structure between target and nontarget categories. We control for stimulus-based differences and find that the earliest EEG signal which correlates with successful categorization of an image is slower, has a variable time of onset, and is postsensory. These results suggest that the timecourse of visual processing may be slower than previously believed, leaving in most cases additional time which visual models might use to implement more time-consuming visual processing steps. The visual system is adept at recognizing not only whole objects, but also the objects it encounters on a daily basis which are only partially visible due to occlusion. Standard feedforward models would suggest that recognition performance on partially visible objects depends only on the visible portions of those objects, while models that build a surface-based representation that the explicit presence of occluders would play an additional role. We investigated recognition performance of partially visible objects in both the presence and absence of occluders and find that the presence of an explicit occluder results in a facilitation of recognition. Differences in the visual processing of occluded and unoccluded partial objects occur quickly and can be seen in the earliest cortical visual areas. These results provide support for surface-based models and suggest that the depth relations in a scene are established early in visual processing.