Learning Fuzzy Membership Functions in a Function-Based Object Recognition System
Learning Fuzzy Membership Functions in a Function-Based Object Recognition System
复制标题
在基于函数的对象识别系统中学习模糊隶属函数
DOI：
10.1007/3-540-58409-9_7
复制
发表时间：
1993
期刊：
影响因子：
--
通讯作者：
K. Bowyer
中科院分区：
文献类型：
--
作者：
K. Woods;D. Cook;L. Hall;L. Stark;K. Bowyer
基于功能的对象识别系统通过推理对象对预期功能的支持程度来识别基本类别级别的对象。这样的系统自然地将“善良的度量”或“成员资格值”与所识别的对象相关联。这种优良性的度量是从物体形状的不同属性的潜在许多原始评估中积累度量的结果。以前，每个原始评估的测量函数都是由识别系统的设计者手工制作的。本文提出了一种自动学习基本评价测度函数的方法，给出了一组用其所需的总体测度标记的示例对象。这里所描述的学习算法应该是普遍适用于任何问题，其中低级别的“模糊”的隶属度值相结合，通过andtree控制结构，以给出一个最终的整体措施。
Functionality-based object recognition systems recognize objects at the basic category level by reasoning about how well they support the expected function. Such systems naturally associate a “measure of goodness” or “membership value” with a recognized object. This measure of goodness is the result of accumulating measures from potentially many primitive evaluations of different properties of the object's shape. Previously the measure function for each of the primitive evaluations has been handcrafted by the designer of the recognition system. A method is presented here for automatically learning the primitive evaluation measure functions given a set of example objects labeled with their desired overall measure. The learning algorithm described here should be generally applicable to any problem in which low-level “fuzzy” membership values are combined through anandtree control structure to give a final overall measure.