De-identification of sensitive data present in images is needed in today’s world, where we have so many frauds and misuse of personal data. Images can communicate a huge variety of personal information. These images have the potential to cause a lot of issues if they are posted on social media, as the affected people’s privacy would be violated by these released images. To protect privacy and meet regulatory obligations in a few cases, it has become necessary to remove or redact any Personally Identifiable Information (PII) wherever possible. Almost all ID cards which an individual possesses contain personally identifiable information, posing privacy risks to individuals. If we could redact just the parts of the image that are PII, this would ensure privacy preservation while retaining the utility of the images.

PII masking is one of the common ways to redact personal information like names, faces, signatures, etc., from an image. In this thesis, we have developed a pipeline to redact just the parts of the ID card containing PII, such that we can still maintain the utility of the ID card. There are many applications of such a mechanism which can redact sensitive information from images and reduce instances of identity theft, which is very common in this digital world. Our framework can be stand-alone and can also be easily integrated with other systems that store or process sensitive images and help strengthen the security and privacy of its users, while still maintaining the utility of the images.