Skip to Main Content
This paper proposes a novel visual learning framework for attention control in active computer vision. The general hierarchical framework is constructed by using reinforcement learning to organize the image processing procedures and find optimal control strategy so as to efficiently reduce the computational cost. This framework allows the interactions between information in different levels and integration of visual modules with other machine learning algorithms, which make it possible to fulfill the specific task quickly by only processing relatively small quantities of data. The experiments of the selective attention on robot are provided to verify the effectiveness of the proposed framework.