Machine learning-based image classification offers a fast and effective way to detect sensitive information within images, allowing for granular real-time DLP policy controls. Our DLP Classification Service accommodates an extensive range of image classifiers, such as passports, driver's licenses, SSNs, checks, payment cards, screenshots, whiteboards, and more. To improve its accuracy and provide better customer experience, we have retrained the classifiers for Release 103.
Significant changes were made to the image classifiers, including:
- A newer underlying model architecture with an 80% increase in the number of parameters
- High-quality labeled training data, including augmented images that are realistic for the image data we see in the cloud, to minimize false positives
According to our lab tests, most of the image classifiers show increased accuracy. If you have any questions or feedback about the image classifiers, please let us know.