Google has made available as open source a machine learning tool that allows users to blur parts of a video or image. The tool can protect people’s identities, the company says.
The tool is called Magritte en is available through GitHub. The machine learning tool can detect objects in videos and images without the need for a powerful computer. If the tool recognizes an object that needs to be blurred, Magritte does this automatically. Examples of objects are according to Google badges and tattoos.
Google says the code is mainly useful for video journalists who want to better guarantee the privacy of data subjects. The tool could fade faster than if you had to do it manually, with still high accuracy, Google claims. Magritte would also be useful for developers and researchers who want to analyze data without being exposed to privacy-sensitive information.
The company also says it has improved its Fully Homomorphic Encryption Transpiler, a technique that the company released a year ago. With fhe, developers can work with encrypted data without having access to privacy-sensitive data in that encrypted data. Google says it has improved that transpiler over the past year; the circuit size is now 50 percent smaller, making the transpiler work faster. According to Google, Fhe would be useful for financial service providers, healthcare and governments, for example, where encrypted data must be used in a secure manner.