In this paper, we propose a novel framework that improves the recognition performance of human support systems, and then discuss why our framework is Human-Centered. A Human-Centered system should have a high recognition ability with minimum burden on the user. Our framework aims to satisfy this requirement by using an artificial agent between a recognition system and the user. If a system is in a difficult situation concerning recognition, an agent will require the user's help. For example, if an object that a system aims to recognize is hidden by the user's hand, the agent will ask the user to move his/her hand. Based on this idea, we implemented a prototype system with two modules: a recognition module to recognize objects and user's motions and an agent module to ask for a user's cooperative action. In the experiment, our prototype system recovers around 50%-70% of the recognition failures caused by three typical difficult situations. The user study reveals that our prototype system has the potential to realize natural and considerate human support systems.