Global and local cues for contour grouping

物体的形状构成了视觉经验的一个中心现象学方面。然而，自然图像通常不包含完整的形状。相反，通常可以确定那些图像中的局部边缘元素的集合。轮廓分组是指确定哪些局部边缘元素属于同一对象边界以及边缘以何种顺序形成边界的过程。以前的轮廓分组方法主要集中在局部属性的边缘对沿着轮廓。在拟议的项目中，我们计划扩展这些方法，将全球性的轮廓属性的影响。这样的全局属性可以是轮廓围绕图像中的闭合区域的事实，或者可以是所得到的形状的整体属性，例如平滑度。我们打算比较计算机算法的预测，执行轮廓分组的人类观察员的表现。通过这种方式，我们希望识别边缘元素集合的特征，这些边缘元素会产生独特而逼真的形状的感知印象。

The shapes of objects form one of the central phenomenological aspects of visual experience. However, natural images typically do not contain complete shapes. Instead it is often possible to determine a collection of local edge elements in those images. Contour grouping refers to the process of deciding which local edge elements belong the same object boundary and in which sequence the edges form the boundary. Previous approaches to contour grouping focused primarily on local properties of edge pairs along a contour. In the proposed project, we plan to extend these approaches to incorporate the effect of global properties of contours. Such global properties might be the fact that a contour surrounds a closed region in an image or might be overall properties of the resulting shape such as smoothness. We intend to compare the predictions from computer algorithms that perform contour grouping to the performance of human observers. This way, we hope to identify the features of a collection of edge elements that give rise to the perceptual impression of a unique and realistic shape.