Nonaccidental properties underlie shape recognition in mammalian and nonmammalian vision
Nonaccidental properties underlie shape recognition in mammalian and nonmammalian vision
复制标题
DOI：
10.1016/j.cub.2006.12.025
复制
发表时间：
2007-02-20
期刊：
影响因子：
9.2
通讯作者：
Wasserman, Edward A.
中科院分区：
文献类型：
--
作者：
Gibson, Brett M.;Lazareva, Olga F.;Wasserman, Edward A.
视网膜上的无限数量的2D图案可以对应于单个3D对象。视觉系统如何解决这个不适定的问题[1]，并在不同的曝光条件下从几个2D视网膜投影中识别物体？目标识别的理论依赖于边缘属性的非偶然统计[2-7]，主要是对称性、共线性、曲线性和共终止性。这些统计数据由成像过程(即3D对象的2D视网膜投影[4])确定；它们在一定范围的视点下的存在使视点不变的识别成为可能。行为生物学中的一个重要问题是，非哺乳动物动物的视觉系统是否也进化了偏见，以利用非偶然统计[8，9]。在这里，我们训练人类和鸽子识别四种形状。使用气泡技术，我们确定了两个物种识别形状所使用的刺激特性。人类和鸽子都使用共同终止，这是现实世界物体最具诊断性的非偶然性质，尽管来自模型计算机观察者的证据表明，共同终止不是这项特定任务中最具诊断性的图画信息。这一结果表明，非哺乳动物的视觉系统在解剖学上不同于人类的视觉系统[11-13]，也偏向于从非偶然统计数据中识别物体。
An infinite number of 2D patterns on the retina can correspond to a single 3D object. How do visual systems resolve this ill-posed problem [1] and recognize objects from only a few 2D retinal projections in varied exposure conditions? Theories of object recognition rely on the nonaccidental statistics of edge properties [2-7], mainly symmetry, collinearity, curvilinearity, and cotermination. These statistics are determined by the image-formation process (i.e., the 2D retinal projection of a 3D object [4]); their existence under a range of viewpoints enables viewpoint-invariant recognition. An important question in behavioral biology is whether the visual systems of nonmammalian animals have also evolved biases to utilize nonaccidental statistics [8, 9]. Here, we trained humans and pigeons to recognize four shapes. With the Bubbles [10] technique, we determined which stimulus properties both species used to recognize the shapes. Both humans and pigeons used cotermination, the most diagnostic nonaccidental property of real-world objects, despite evidence from a model computer observer that cotermination was not the most diagnostic pictorial information in this particular task. This result reveals that a nonmammalian visual system that is different anatomically from the human visual system [11-13] is also biased to recognize objects from nonaccidental statistics.