Fakecatcher: detection of synthetic portrait videos using biological signals

A non-transitory computer readable medium storing instructions for controlling a programmable processor to analyze a received video comprising skin for indicia of synthetic origin, comprising:

• instructions for generating signal maps of the skin;
• instructions for implementing a network selected from the group consisting of a convolutional neural network and a support vector network for determining a spatial coherence and temporal consistency of biological signals represented in the video of the skin dependent on a power spectral density of respective signal maps;
• and
• instructions for producing an output dependent on the determined spatial and temporal consistency of the biological signals.

A system for analyzing a video comprising skin for indicia of synthetic origin, comprising:

• an input port configured to receive the video;
• at least one automated processor responsive to a spatial coherence and temporal consistency of biological signals represented in the video of the skin, configured to:
• generate signal maps of the skin;
• extract feature vectors from the video comprising biological signals dependent on the generated signal maps;
• classify the feature vectors with at probabilistic video classifier selected from the group consisting of a convolutional neural network and a support vector network, dependent on a power spectral density of respective signal maps;
• and
• determine a likelihood of synthetic origin with a probabilistic video classifier;
• and
• an output port configured to communicate a signal selectively responsive to a probability of synthetic origin of the video, dependent on the determined spatial and temporal consistency.