Iterative Models in Figure-Ground Perception: Tests and Challenges

视觉感知似乎很简单。尽管如此，长期以来，它经受住了计算机科学家和视觉科学家试图破解其密码的努力，很可能是因为过去的理论主要基于有意识的视觉。为了让感知者体验到一个连贯的视觉世界，感知输入必须被组织成单独的物体或“图形”。作为这个过程的一部分，大脑决定一个图形相对于两个连续空间区域之间的每个边界的位置。以垂直边框为例。大脑决定这个边界是一个人躺在左边还是右边的边界。当边界被认为是一个躺在左边的人物的边界时，右边的区域似乎只是继续在边界上的人物后面；那里没有形状。这些“图形背景”的决定是在意识意识之外做出的。正因为如此，研究产生图形-背景感知的机制是极其困难的。一个重要的，尚未解决的问题是，图形-背景感知是通过快速的前馈机制完成的，还是涉及来自更高处理水平的反馈的迭代机制。在前馈模型中，输入在连续的阶段被处理，直到一个连贯的感知出现。在迭代模型中，前馈处理是不够的；从高层到低层的反馈对于创造感知是必要的。亚利桑那大学的玛丽·彼得森（Mary Peterson）和她的同事们设计了视觉显示器，让他们能够调查这个问题。一种显示类型是一个小的对称的、有界的剪影躺在一个较大的地面上。部分熟悉的形状隐藏在轮廓边缘的地面上；这些形状不是有意识地感知到的，只有封闭轮廓的形状是有意识地感知到的。（经典的“脸花瓶错觉”就是一个例子。）先前的实验表明，当被隐藏物体的形状不被感知时，它的形状被抑制了，这支持了图形-背景感知是由可能在边界两侧看到的形状之间的抑制竞争产生的观点；胜利者被认为是有形状的人物，而失败者则被压制，而可能被看到的那部分空间被认为是没有形状的地面。这些显示器的设计是为了在形状处理阶段隔离竞争。彼得森博士和他的同事们还将进行一系列实验，以测试是否可以在较低的层次上观察到抑制，在较低的层次上观察到单个部分的表现，在较高的层次上观察到形状描述（“语义”）的表现。根据前馈解释，对丢失形状的抑制将阻止对其语义的访问；因此，在较高的水平上没有明显的影响。如果没有反馈，在较低的部分加工水平上也不会有明显的效果。迭代观点可以通过假设形状之间竞争的结果传递到更高和更低的级别来解释较低和/或较高级别的抑制。因此，提出的实验将在这些关于感知如何发生的相互竞争的观点之间做出裁决。研究人员还将试图确定在拥挤的现实场景中将人物与场地分开的过程。这些更复杂显示的感知构成了对当前图形-背景感知迭代模型的挑战。计划中的研究将为视觉的神经生理学实验和正式计算模型提供基础，并将有助于我们对形状感知的时空动态的理解。

Visual perception seems trivially easy. Nevertheless, it has long withstood the attempts of both computer scientists and vision scientists to crack its code, most likely because past theories were based predominantly on conscious vision. In order for perceivers to experience a coherent visual world, perceptual input must be organized into separate objects, or "figures." As part of this process, the brain decides where a figure lies with respect to every border between two contiguous regions of space. Consider a vertical border, for instance. The brain decides whether that border is a boundary for a figure lying on the left or the right. When the border is perceived as a boundary of a figure lying on the left, the region on the right seems simply to continue behind the figure at the border; no shape is perceived there. These "figure-ground" decisions are made outside of conscious awareness. Because of this, it is extremely difficult to investigate the mechanisms that produce figure-ground perception. An important, unanswered, question is whether figure-ground perception is accomplished via fast, feed-forward mechanisms or whether iterative mechanisms involving feedback from higher processing levels are involved. In feed-forward models, the input is processed in successive stages until a coherent percept emerges. In iterative models, feed-forward processing is not sufficient; feedback from higher to lower levels is necessary to create the percept.Mary Peterson and her colleagues at the University of Arizona have designed visual displays that allow them to investigate this question. One display type is a small symmetric, bounded silhouette lying on a larger ground. Portions of familiar shapes are hidden along the groundside of the silhouette's border; these shapes are not perceived consciously, only the shape of the enclosed silhouette is perceived consciously. (The classic "face-vase illusion" is an example of this.) Previous experiments have shown that the shape of the hidden object is suppressed when it is not perceived, supporting the view that figure-ground perception results from inhibitory competition between shapes that might be seen on opposite sides of a border; the winner is perceived as the shaped figure, whereas the loser is suppressed and the portion of space where it might have been seen is perceived as a shapeless ground. These displays are designed to isolate competition at the shape processing stage. Dr. Peterson and colleagues will also conduct a series of experiments which test whether suppression can be observed at lower levels where individual parts are represented and at higher levels where shape descriptions ("semantics") are represented. According to a feed-forward account, suppression of the losing shape would prevent access to its semantics; hence, no effects should be evident at higher levels. Without feedback, no effects should be evident at lower, part-processing, levels either. An iterative view could account for suppression at lower and/or higher levels by assuming that the outcome of the between-shape competition was relayed to higher and lower levels. Thus the proposed experiments will adjudicate between these competing views of how perception occurs. The researchers will also attempt to identify the processes involved in segregating figures from grounds in crowded real-world scenes. The perception of these more complex displays constitutes a challenge to current iterative models of figure-ground perception. The planned research will provide a foundation for neurophysiological experiments and formal computational models of vision and will contribute to our understanding of the temporal and spatial dynamics of shape perception.