EE ZOOM Seminar: Neural Alignment for Face De-pixelization
השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בצ'אט
Join Zoom Meeting
https://us04web.zoom.us/j/6207215556
Speaker: Maayan Shuvi
M.Sc. student under the supervision of Prof. Daniel Cohen-Or
Wednesday, April 22nd, 2020 at 14:00
ZOOM
Neural Alignment for Face De-pixelization
Abstract
We present a method to reconstruct a high-resolution video from a face-video, where a person’s identity was obscured by pixelization. This hiding method is popular because the viewer can still perceive a human face figure and the overall head motion. However, we show in our experiments how a fairly good approximation of the original video is reconstructed so that anonymity is compromised.
Our system exploits the simultaneous similarity and small disparity between close-by video frames depicting a human face, and employs a spatial transformation component that learns the alignment between the pixelated frames.
Each frame, supported by its aligned surrounding frames, is first encoded, then decoded to a higher resolution. Reconstruction and perceptual losses promote adherence to the ground-truth, and an adversarial loss assists in maintaining domain faithfulness. There is no need for explicit temporal coherency loss as it is maintained implicitly by the alignment of neighboring frames and reconstruction.
Our work shows that given the specific prior of a human face, which is well structured, and using multiple aligned frames, there is enough information in a pixelated video stream to reconstruct a high-quality approximation of the original signal.
