一种引入注意机制的视觉计算模型

提出了一种基于注意机制的视觉模型，其特点是：将注意过程分为３个层次，分别模拟生物的瞳孔聚焦、眼动和头动；提出一种新的可变结构的非均匀采样映射来模拟生物视网膜特性；强调了数据驱动的ｂｏｔｏｍ－ｕｐ过程与知识驱动的ｔｏｐ－ｄｏｗｎ过程的融合；提出新颖的类似树形的知识表示方法和基于深度优先搜索树的注意点转移控制机制。

A Visual Computing Model Based on Attention Mechanism

A visual computing model based on attention mechanism with the following characters is proposed in this paper.First, it imitates the characteristic of zoom-lens, eye movement and head movement of human vision system. Second, the retina character is modeled by a new nonuniform sampling method with changeable structure. Third, the bottom-up process driven by data and top-down process driven by knowledge are combined. A novel knowledge representation method and a fixation movement controlling mechanism based on depth first search tree are proposed as well.