Because, we want to protect the image better. Other typical places to add information to an image such as metadata (the primary method fueling C2PA), are all too easy removed when an image is shared or otherwise processed.
Does the image get distorted by encoding a Signature?
While the image is slightly altered as part of embedding the signature, it is done in a way that only minimally interferes with human perception. Technically speaking, embedding a signature typically results in a PSNR loss 47 which is equivalent to saving a JPEG image at 90% quality.
Is the Signature the same as a Digital Watermark?
What we've described is very similar to classical digital watermarks, but there's more to Sasha Signatures than that. Outside of ensuring that every Signature is unique, we also tie other unique data to the particular image at hand, making it extremely hard to move a signature from one image to another.
Can the signature be traced back to the image content?
No, all that is embedded into the image is a unique randomized numeric value. This identifier can not be traced to image content, or to who added the Signature, or to where it was done.
Can a signature be forged?
No, we don’t believe so: we’ve gone to great lengths testing against “adversarial attacks” where a signature is applied to another image. In each case so far, we’ve been able to detect it.
What information, exactly, is embedded into the image?
All that is embedded into the image is a unique, randomized, numeric value. This identifier doesn't contain anything that can be traced to the image contents, or who added the Signature, where it was done, etc. – it is only a unique identifier tied to that particular image.
Can personal information be disclosed?
No. The individual user or company can choose to associate data that, depending on use context, can help prove ownership and document who the image was shared with. But it does not contain personally identifiable information
