Systems and methods for distorting captcha images with generative adversarial networks

A method, comprising:

• storing image records, each image record comprising an image field, a distorted image field, a label field and an applied algorithm identifier field;
• training a machine learning model that comprises a plurality of adversarial attack algorithms, each adversarial attack algorithm having a unique identifier;
• selecting one of the image records;
• retrieving a selected image stored in the image field of the selected image record;
• determining a size of the selected image;
• applying one of the adversarial attack algorithms to create an array of pixels that is the size of the selected image, wherein a majority of the pixels in the array are zero values and the remaining pixels in the array are noise-level red-green-blue values;
• merging the array of pixels with the selected image to form a distorted image;
• updating the selected image record by storing the distorted image in the distorted image field of the selected image record and storing the unique identifier of the applied adversarial attack algorithm in the applied algorithm identifier field of the selected image record;
• selecting only a set of image records that have a same label stored in the label field;
• retrieving a showcase set of distorted images from a plurality of distorted image fields of the selected set of image records;
• and
• providing a CAPTCHA test using the showcase set of distorted images, by displaying only the set of distorted images with the same label.

A non-transitory computer-accessible medium having stored thereon computer-executable instructions, wherein upon execution by a processor, the instructions cause the processor to perform a method comprising:

• storing image records, each image record comprising an image field, a distorted image field, a label field and an applied algorithm identifier field;
• training a generative adversarial network that generates a plurality of adversarial attack algorithms, each adversarial attack algorithm having a unique identifier;
• selecting one of the image records;
• retrieving a selected image stored in the image field of the selected image record;
• determining a size of the selected image;
• applying one of the adversarial attack algorithms to create an array of pixels that is the size of the selected image;
• merging the array of pixels with the selected image to form a distorted image;
• updating the selected image record by storing the distorted image in the distorted image field of the selected image record and storing the unique identifier of the applied adversarial attack algorithm in the applied algorithm identifier field of the selected image record;
• refining a selected set of image records to those that have different values in the applied algorithm identifier field;
• selecting only a set of image records that have a same label stored in the label field;
• retrieving a showcase set of distorted images from a plurality of distorted image fields of the selected set of image records;
• and
• providing a CAPTCHA test using the showcase set of distorted images, by displaying only the set of distorted images with the same label.