The Perceptual Identification and Representation of Image Contours

人们很早就认识到，一个物体的令人信服的图形表示有时可以通过在图像中画出几个显著的轮廓来实现。这种现象真的很了不起，因为线条画有效地去除了自然场景中通常可用的几乎所有颜色和阴影的变化。不知何故，创作这种图画的艺术家能够通过几笔简单的笔触捕捉到感知识别的基本信息。尽管一幅结构良好的线条画很容易被人类观察者解读，但创作这些图画的能力可能需要相当高的艺术技能。事实上，尽管在计算机视觉领域进行了近半个世纪的研究，但现有的算法无法复制称职的人类艺术家的表现。在这个项目中，詹姆斯·托德博士和他在俄亥俄州立大学的学生将研究人类观察者如何感知地识别不同类型的图像轮廓，如阴影、角落或遮挡。该小组还将研究图像中的哪些轮廓在感知上对创建物体的图画表示最重要。这些研究中的刺激将包括接受过不同程度培训的艺术家的素描，他们将被要求绘制具有已知3D结构的物体的线条图。人类观察者将对这些图画进行排序，以评估它们的相对感知效果。图中的轮廓也将与所描绘的表面几何的不同方面进行比较，以确定表面的哪些特定方面对于其图画描绘是最重要的。更好地理解人类如何从2D图像数据中感知地确定表面的3D形状具有许多可能的应用，包括设计更健壮和有效的机器视觉算法，改进计算机图形学和设计中的3D可视化技术，以及为盲人开发更多功能的假肢设备。这项工作也可能对如何在艺术或设计课程中教授学生绘画产生重大影响。

It has long been recognized that a convincing pictorial representation of an object can sometimes be achieved by drawing just a few salient contours in an image. This phenomenon is really quite remarkable, given that a line drawing effectively strips away almost all of the variations in color and shading that are ordinarily available in natural scenes. Somehow the artists who create such drawings are able to capture the essential information for perceptual recognition with just a few simple strokes. Although a well structured line drawing is easily interpreted by human observers, the ability to create these drawings can require considerable artistic skill. Indeed, despite almost a half century of research in the field of computer vision, there are no existing algorithms that can duplicate the performance of a competent human artist. In this project, Dr. James Todd and his students at the Ohio State University will investigate how human observers perceptually identify different types of image contours, such as shadows, corners or occlusion. The group will also examine which contours in an image are perceptually most important for creating pictorial representations of objects. The stimuli in these studies will include drawings by artists with varying amounts of training, who will be asked to produce line drawings of objects with known 3D structures. The drawings will be ranked by human observers to assess their relative perceptual effectiveness. The contours in the drawings will also be compared with different aspects of the depicted surface geometry in order to determine which specific aspects of a surface are most important for its pictorial depiction. A better understanding of how human observers perceptually determine the 3D shapes of surfaces from 2D image data has many possible applications, including the design of more robust and effective algorithms in machine vision, improved techniques for 3D visualization in computer graphics and design, and the potential development of more functional prosthetic devices for the blind. This work may also have a significant impact on how students are taught to draw in art or design courses.