Face blurring, in video sequences, is an available option to disguise the identity of the person seen in the film.
Abstract
Face blurring, in video sequences, is an available option to disguise the identity of the person seen in the film.
Pixelation blur is used in news items to prevent identification and harming the film subject. On the other hand, such blur can also be used by criminal, filming himself, and trying not to be identified.
This work handles video sequences that were blurred by pixelation, and examines the possibility to reconstruct the original face image from the blurred movie. The reconstruction algorithm presented in this project is using super resolution methods, dictionary image reconstructing (using K-SVD algorithm) and deconvulution.
A novel motion estimation method is presented here, to estimate the motion in a degraded resolution video sequence.
The algorithm used here gives the best results for blurred films that has high original resolution, and movement only in the screen plane.
The algorithm was tested on several examples.
Authentic videos reconstruction has low quality. We will discuss the algorithm expansions, needed to improve the results, and the video types that can provide relatively high reconstruction quality.
This algorithm can be used for blurring quality assurance, or to identify details that were blurred in a video sequence.
Results
