Detecting and monitoring unauthorized uses of visual content

A non-transitory computer-readable storage medium having stored thereon computer executable instructions, which when executed by a computing device, cause the computing device to be operable for:

• receiving query content that includes first visual data;
• selecting reference content that includes second visual data;
• determining a model, wherein the model comprises:
• a first sub-model that is trained with a first training dataset to output a first score based on a comparison of a content pair to detect visible alterations in the content pair, and
• a second sub-model that is trained with a second training dataset to output a second score based on a comparison of a content pair for a visible match as distinguishable by a human wherein the second sub-model outputs a higher second score for similarity when visible alterations are visually indistinguishable to a human compared to visually identifiable, wherein the first sub-model is trained to output a higher score for similarity than the second sub-model when visible alterations arefound in the content pair and are identifiable by the human;
• generating a first visual feature mapping for the query content using the first sub-model and the second sub-model of the model;
• determining a second visual feature mapping for the reference content, wherein the second visual feature mapping is determined using the first sub-model and the second sub-model of the model;
• comparing the first visual feature mapping and the second visual feature mapping for the first sub-model and the second sub-model to generate a third score that rates a similarity between the query content and the reference content;
• and
• outputting the third score.

An apparatus comprising:

• one or more computer processors;
• and
• a computer-readable storage medium comprising instructions for controlling the one or more computer processors to be operable for:
• receiving query content that includes first visual data;
• selecting reference content that includes second visual data;
• determining a model, wherein the model comprises:
• a first sub-model that is trained with a first training dataset to output a first score based on a comparison of a content pair to detect visible alterations in the content pair;
• and
• a second sub-model that is trained with a second training dataset to output a second score based on a comparison of a content pair for a visible match as distinguishable by a human wherein the second sub-model outputs a higher second score for similarity when visible alterations are visually indistinguishable to a human compared to visually identifiable, wherein the first sub-model is trained to output a higher score for similarity than the second sub-model when visible alterations are found in the content pair and are identifiable by the human;
• generating a first visual feature mapping for the query content using the first sub-model and the second sub-model of the model;
• determining a second visual feature mapping for the reference content, wherein the second visual feature mapping is determined using the first sub-model and the second sub-model of the model;
• comparing the first visual feature mapping and the second visual feature mapping for the first sub-model and the second sub-model to generate a third score that rates a similarity between the query content and the reference content;
• and
• outputting the third score.