Determination of high risk images using peer-based review and machine learning

A system for image arbitration comprising at least one processor configured to:

• generate one or more profiles each associated with a corresponding first user of a networked system and pertaining to problematic conditions for image capture, wherein the one or more profiles are generated in part based on classification of a plurality of images into respective risk categories by crowdsourcing users of the networked system;
• capture an image of a second user via an image capture device;
• upload the captured image of the second user to the networked system;
• compare the captured image of the second user to the generated one or more profiles to determine a score value for the captured image of the second user, wherein the score value is a combination of a metadata score value, a geolocation score value, and an image score value based at least in part on a perspective of the captured image;
• classifying, based upon the score value, the captured image into one of a plurality of risk categories using a latent class model;
• in response to the classification:
• reject upload of the captured image of the second user in response to the classification of the captured image based at least in part on the score value satisfying a first threshold associated with a risk category indicating a presence of problematic conditions during capturing of the image of the second user;
• allow upload of the captured image of the second user, via the processor, in response to classification of the captured image based at least in part on the score value satisfying a second threshold associated with a risk category indicating an absence of problematic conditions during capturing of the image of the second user;
• and
• provide, when the score value falls in between the first threshold and the second threshold, the captured image to a peer group to determine whether to classify the captured image into a risk category indicating a presence of problematic conditions or a risk category indicating an absence of problematic conditions.

A computer program product for image arbitration, the computer program product comprising one or more computer readable storage media collectively having program instructions embodied therewith, the program instructions executable by a processor to:

• generate one or more profiles each associated with a corresponding first user of a networked system and pertaining to problematic conditions for image capture, wherein the one or more profiles are generated in part based on classification of a plurality of images into respective risk categories by crowdsourcing users of the networked system;
• capture an image of a second user via an image capture device;
• upload the captured image of the second user to the networked system;
• compare the captured image of the second user to the generated one or more profiles to determine a score value for the captured image of the second user, wherein the score value is a combination of a metadata score value, a geolocation score value, and an image score value based at least in part on a perspective of the captured image;
• classifying, based upon the score value, the captured image into one of a plurality of risk categories using a latent class model;
• in response to the classification:
• reject upload of the captured image of the second user in response to the classification of the captured image based at least in part on the score value satisfying a first threshold associated with a risk category indicating a presence of problematic conditions during capturing of the image of the second user;
• allow upload of the captured image of the second user, via the processor, in response to classification of the captured image based at least in part on the score value satisfying a second threshold associated with a risk category indicating an absence of problematic conditions during capturing of the image of the second user;
• and
• provide, when the score value falls in between the first threshold and the second threshold, the captured image to a peer group to determine whether to classify the captured image into a risk category indicating a presence of problematic conditions or a risk category indicating an absence of problematic conditions.