Disclosed is detecting screenshot images and protecting against loss of sensitive screenshot image data. The method collects examples of screenshot images and non-screenshot images, creates labelled ground-truth data for the examples, and applies re-rendering of collected images to represent variations that may contain sensitive information. The method includes training a DL stack by forward inference and back propagation using labelled ground truth data for the screenshots, storing parameters of the trained DL stack for inference from production images, and using a production DL stack with the stored parameters to classify at least one production image by inference as containing a screenshot image. Further, DL stack includes a first set of layers closer to an input layer being pre-trained to perform image recognition before exposing a second set of layers further from the input layer of DL stack to the labelled ground truth data for the screenshot images and non-screenshot images.