To perceive an object in a scene, the visual system must integrate local image elements together for a coherent percept. However, in any sufficiently complex scene, there are multiple possible ways to organize local elements. Hence, it is a challenge for the visual system to find the right correspondence among local elements. For instance, to perceive symmetry, the visual system needs to find correspondence between image elements across a symmetry axis. However, if the location and orientation of the symmetry axis are unknown, the midpoint between any pairs of image elements is a candidate for a symmetry axis. We measured symmetry detection under various contexts and different amounts of axis-orientation uncertainty. Our result was best described by a multiple-channel model in which each channel tunes to a specific axis orientation. The response of each channel is the number of corresponding elements consistent with the tuned symmetry axis divided by an inhibition signal from other channels. Similar results and computation principles are also found in contour integration, Glass pattern perception, and Ebbinghaus size illusion. Thus, divisive inhibition, which was originally proposed to explain phenomena in the contrast domain, is ubiquitous in perceptual grouping. It serves to suppress unwanted groupings and to ensure the emergence of the right ones.

In two neuroimaging studies, we aim to examine the role of the hippocampus cortex (HPC) in binding and differentiating during encoding and retrieval of face-scene composite images. Two neuroimaging experiments using fMRI were conducted. In Study 1, we examined to role of HPC in binding during encoding and retrieval of face-scene composites. During encoding, participants were asked to perform a 2-back working memory (WM) task. For face-scene composite images, they made an affirmative response only when the currently displayed image, in terms of both face and scene, was identical to the one presented two images back. As a control, participants were asked to undertake the same WM task with face-only and scene- only images. During recognition, they were shown face-scene composite images and judged whether their combination was identical to that during encoding and indicated confidence in their judgments. When they were not identical, the composite images were recombination of those that were shown during encoding. Results showed that, in contrast to face-only and scene-only images, both the left and right HPC exhibited greater activation during encoding of face-scene composites. Likewise, for composite images, bilateral HPC exhibited greater activation when we contrasted between those that were exact or partial repetition versus those that were not repeated. During recognition, bilateral anterior HPC showed greater activation when the recombined composite images contrasted with the old composites. Moreover, bilateral HPC showed greater activation when the recombined composites were correctly rejected than when they were erroneously identified (i.e., false alarm, FA), which may explain the relatively high correct rejection (CR) rate in judging the recombined composite images. In Study 2, we examined the neural mechanisms that may underlie the retrieval of episodic memory for face-scene composite images of differential associative strength. During encoding, participants were shown face-scene composite images where a specific face was associated with a specific scene, a single face associated with multiple scenes, or multiple repetitions of a specific pairing between a face and a scene. During recognition, participants were shown face-scene images and asked to judge whether the specific pairing of face and scene was presented during encoding or was a re-combined version from those presented during encoding. They performed the recognition test while their brains were scanned with fMRI. The contrast between Fan 1-1 and Fan 1-5 highlights stronger activations of bilateral FFA to differentiate the specific links between a face and multiple scenes. On the other hand, the contrast between Fan 1-1 and its counterpart of five repetitions (i.e., R5) revealed all the relevant brain regions were more strongly activated because of multiple encounters. Finally, and somewhat unexpectedly, the activations of bilateral HPC were diminished in the contrast between R5 and Fan 1-5, suggesting the dual (and counteracting) role of HPC for binding in the former case and pattern differentiation in the latter case. Taken together, the findings of the present studies implicate the role of HPC in binding and differentiating face-scene composites when participants intentionally encode those composites into and retrieve them from episodic memory.