Visual computation of surfaces under partial occlusion and transparency

在国家科学基金会的支持下，曼尼什·辛格博士将进行为期两年的视觉感知研究--我们如何看待一个由重叠物体组成的复杂、不断变化的世界。感知的世界包括三维物体和表面。然而，我们眼睛的输入不是“物体”或“表面”，而是投射到每个视网膜上的光的二维阵列的波动。随着我们在这个世界上的移动，这些视网膜阵列发生了很大的变化--当我们穿过房间时，沙发部分隐藏在咖啡桌后面，或者狗部分被隐藏在茂密的树叶后面。视觉的一个基本且尚未解决的问题是理解我们的大脑如何将如此复杂的变化的光阵列分类为可感知的表面和物体。这项由计算机生成的立体显示器使用的是一些物体被其他物体部分遮挡的立体显示器，因此只向眼睛呈现其边界的不连续部分。尽管如此，观察者感觉到的是完整的、单一的物体--而不是不同的碎片。研究程序的一个方面检查涉及表面形状和轮廓几何的约束，这些约束可以确定部分遮挡对象的连续性和形状。这项研究的第二个方面涉及遮挡对象何时是部分透明的。在这种情况下，从两个不同表面(透明表面和底层不透明表面)反射的光在视网膜处合并，但我们看到两个独立的表面沿同一视线放置在不同的深度。这项研究的组成部分调查了大脑用来将图像亮度分割成两个不同表面的几何(图形)和光度(光强度)限制。这项研究将增强对人类视觉中心方面的理解。此外，它还将提供有价值的知识，可用于构建人工视觉系统，从相机图像中恢复表面布局。

With National Science Foundation support, Dr. Manish Singh will conduct two years of research on visual perception-how we see a complex, changing world of overlapping objects. The perceived world includes three dimensional objects and surfaces. The inputs to our eyes, however, are not "objects" or "surfaces," but fluctuations in the two-dimensional arrays of light projected onto each retina. These retinal arrays change a great deal as we move in the world-as a couch gets partly hidden behind a coffee table when walking across a room, or a dog is partly seen behind thick foliage. A fundamental and unsolved problem of vision is to understand how our brain can sort out such complex changing light arrays into perceived surfaces and objects.The funded research uses computer-generated stereoscopic displays in which some objects are partly occluded by other objects, and thus present only disjoined portions of their boundaries to the eyes. Despite this, observers perceive complete, unitary, objects-not disparate fragments. One aspect of the research program examines constraints involving both surface shape and contour geometry that may determine the continuity and shape of partly occluded objects. A second aspect of this research concerns when an occluding object is partially transparent. In this case, light reflected from two different surfaces (the transparent, and the underlying opaque surface) is merged at the retinas, but we see two separate surfaces placed at different depths along the same line of sight. This component of the research investigates the geometric (figural) and photometric (light intensity) constraints that the brain uses to split image luminance into two distinct surfaces placed at different depths.This research will enhance understanding of a central aspect of human vision. Additionally, it will contribute valuable knowledge that can be used to construct artificial vision systems for recovery of surface layouts from camera images.