Visual Digest Networks

摘要

Attention, understanding and abstraction are three key elements in our visual communication that we have taken for granted. These interconnected elements constitute a Visual Digest Network. In this chapter, we investigate the conceptual design of Visual Digest Networks at three visual abstraction levels: gaze, object and word. The goal is to minimize the media footprint during visual communication while sustaining essential semantic communication. The Attentive Video Network is designed to detect the operator's gaze and adjust the video resolution at the sensor side across the network. Our results show significant improvements in network bandwidth utilization. The Object Video Network is designed for mobile video network applications, where faces and cars are detected. The multi-resolution profiles are configured for media according to the network footprint. The video is sent across the network with multiple resolutions and metadata; controlled by the bandwidth regulator. The results show that the videorn can be transmitted in the low-bandwidth conditions. Finally, the Image-Word Search Network is designed for face reconstruction across the network. In this study, we assume the hidden layer between the facial features and referral expressive words contain 'control points' that can be articulated mathematically, visually and verbally. This experiment is a crude model of the semantic network. Nevertheless, we see the potential of the two-way mapping.