An Affect-Responsive Interactive Photo Frame

We develop an interactive photo-frame system in which a series of videos of a single person are automatically segmented and a response logic is derived to interact with the user in real-time. The system is composed of five modules. The first module analyzes the uploaded videos and prepares segments for interactive play, in an offline manner. The second module uses multi-modal input (activity levels, facial expression, etc.) to generate a user state. These states are used by the internal frame logic, the third module, to select segments from the offlinegenerated segment dictionary, and they determine the response of the system. A continuous video stream is synthesized from the prepared segments in accordance with the modeled state of the user. The system logic includes online/offline adaptation, which is based on stored inputoutput pairs during real-time operation, and offline learning to improve the system response. The fourth module is the application interface, which deals with handling the input and output streams. Finally, a dual-frame module is described to enhance the use of the system.