A Formal Model of Visual Feature Integration

9319103 IVRY在心理学，生理学，甚至计算机科学中的对象识别理论都与特征分析有关，其中特征可能是线段，颜色，空间频率等。根据这些理论，识别一个物体包括检测该物体的特征。例如，圆形、闪亮和红色的特征可能表示苹果。然而，最近，Treisman和她的同事们指出，正确的特征配准不足以进行真实的目标识别。当许多不同的物体存在时，就像它们在大多数自然场景中一样，观察者不仅要正确地注册特征，还要正确地将特征绑定到单独的物体中。Treisman和其他人已经证明，在短暂暴露的条件下，观察者会报告“虚幻的连接”，或者他们正确识别视觉特征，但错误地将它们组合在一起的戒律。例如，在某些实验室条件下，一个人短暂地看到一个红色的X和一个绿色的O，有多达30%的人会认为X是绿色的。在大多数情况下，我们的视觉系统结合物体的特征似乎没有错误或努力，使得特征整合的过程很难研究。通过研究在实验室条件下特征绑定过程中的误差（即虚幻连词），我们可以检验不同的特征整合理论。不幸的是，由于三个原因，特性绑定的研究受到了阻碍。首先，大多数理论都是以模糊或非正式的方式加以说明的。其次，理论还没有形成一种允许它们之间进行直接检验的形式。最后，在如何测量虚幻连词的问题上存在相当大的分歧。许多广泛使用的从虚幻连词的出现来测量特征绑定的方法将特征配准与特征绑定混淆。初步的研究已经导致了一个数学模型的发展，该模型解决了上述三个问题，并基于概率多维建模和位置感知的心理物理学，发展了一个新的特征绑定理论。将进行的15个实验分为四类：(1)形式理论将通过系统地改变刺激值（如项目的颜色）来检验，并观察它们如何影响数学模型的参数，必要时调整模型。(2)注意力等认知变量会发生变化，这些变量对特征整合的影响是单独估计的。(3)直接比较不同的特征绑定理论。(4)分析方法和形式分析将扩展到新的实验范式，以使其尽可能对其他研究者有用。＊＊＊

9319103 IVRY Theories of object recognition in psychology, physiology, and even computer science have concerned themselves with feature analysis, where features may be line segments, colors, spatial frequencies, etc. According to these theories, recognizing an object involves detecting the features of that object. For example, the features round, shiny, and red might signal an apple. Recently, however, Treisman and her colleagues have pointed out that correct feature registration is not sufficient for veridical object recognition. When many different objects are present, as they are in most natural scenes, observers must not only correctly register the features, but also correctly bind the features into separate objects. Treisman and other have shown that under conditions of a brief exposure, observers will report "illusory conjunctions," or precepts in which they correctly identify the visual features, but combine them incorrectly. For example, an individual briefly presented a red X and a green O would perceive the X as green in some laboratory conditions on up to 30% of the trials. In most circumstances, our visual systems combine features of objects seemingly without error or effort, making the process of feature integration difficult to study. By studying errors in this feature binding process under laboratory conditions (i.e., illusory conjunctions), we can test different theories of feature integration. Unfortunately, the study of feature binding has been hampered for three reasons. First, most theories have been specified in a vague or informal manner. Second, theories have not been formulated in a format that allows a direct test between them. Finally, there has been considerable disagreement on how to measure illusory conjunctions. Many of the widely used methods of measuring feature binding from the occurrence of illusory conjunctions confound feature registration with feature binding. Preliminary research has led to the development of a mathematical model that add resses the above three issues, as well as to the development of a new theory of feature binding, based on probabilistic multidimensional modeling and on the psychophysics of location perception. The 15 experiments to be performed fall into four categories: (1) The formal theory will be tested by varying stimulus values (e.g., color of items) systematically and seeing how they affect parameters of the mathematical model, adjusting the model if necessary. (2) Cognitive variables such as attention will be varied and the effect of these variables on feature integration estimated separately. (3) Different theories of feature binding will be compared directly. (4) The analytic method and formal analysis will be extended to new experimental paradigms, in order to make it as useful as possible to other investigators. ***