In this paper, we propose to model the edge information contained in natural scenes as points in the 3D space of positions and orientations. This space is equipped with a strong geometrical structure and it is identified as the rototranslation group. In this space, we compute a histogram of co-occurrence of edges from a database of natural images and show that it can be interpreted as a probability density function, expressed by the fundamental solution of a suitable Fokkera??Planck equation defined in the 3D structured space. Both estimated statistics and model predictions are reconsidered and compared with the partial gestalt association fields proposed by D. J. Field, A. Hayes, and R. F. Hess (1993). Finally, parametric identification allows to estimate the variance of the co-occurrence random process in natural images.
展开▼
机译：在本文中，我们建议将自然场景中包含的边缘信息模拟为位置和方向的3D空间中的点。该空间配有强大的几何结构，并将其识别为旋转转换组。在这个空间中，我们计算来自自然图像数据库的边缘的共同发生的直方图，并显示它可以被解释为概率密度函数，由3D中定义的合适Fokkera的基本解决方案的基本解决方案表示结构化空间。估计的统计数据和模型预测都被重新考虑并与D. J. Field，A. Hayes和R. F. Hess（1993）提出的部分格式塔基协会领域进行了重新考虑。最后，参数识别允许估计自然图像中共发生随机过程的方差。
展开▼
