Generating a challenge-response for authentication using relations among objects

An apparatus comprising:

• a database configured to store ground-truth images;
• one or more processors;
• and
• memory storing instructions that, when executed by the one or more processors, cause the apparatus to:
• extract a plurality of objects from one or more ground-truth images;
• cluster, based on probabilities of co-occurrence of each respective object of the plurality of objects with other objects of the plurality of objects in the one or more ground-truth images, the plurality of objects;
• generate an initial captcha image from the plurality of objects, wherein the initial captcha image comprises a first object and a second object;
• generate a synthetic captcha image from the plurality of objects by replacing the first object in the initial captcha image with a third object of the plurality of objects;
• transmit, to a user device, at least two challenge images as a security challenge, wherein the at least two challenge images comprise the initial captcha image and the synthetic captcha image;
• and
• determine, based on a response to the security challenge, whether to authorize a user access request by:
• determining a first relatedness of the first object and the second object based on the clustering;
• determining a second relatedness of the third object and the second object based on the clustering;
• comparing the first relatedness and the second relatedness;
• and
• determining, based on the comparing and the response to the security challenge, whether to authorize the user access.

One or more non-transitory media storing instructions that, when executed by one or more processors, cause the one or more processors to perform steps comprising:

• extracting a plurality of objects from one or more ground-truth images;
• clustering, based on probabilities of co-occurrence of each respective object of the plurality of objects with other objects of the plurality of objects in the one or more ground-truth images, the plurality of objects;
• generating a plurality of first captcha images using a generative adversarial network (GAN) model, wherein the GAN model has a GAN architecture comprising a generator and a discriminator;
• generating, using the GAN model, a plurality of second captcha images by replacing a first set of objects in the plurality of first captcha images with a second set of objects;
• transmitting, to a user device, at least two challenge images as a security challenge, wherein the at least two challenge images comprise a first captcha image from the plurality of first captcha images and a second captcha image from the plurality of second captcha images;
• and
• determining, based on a response to the security challenge, whether to authorize a user access request.