Google Releases Research On How Easy It Is To Remove Watermarks From Images

Google, on its research blog, revealed details on how easy it is for a computer algorithm to bypass standard photo watermarking practices and remove watermark protection. The end result is that these images are for reuse across the internet without giving the original source any credit.

Presented at 2017 Computer Vision and Pattern Recognition Conference, the research titled “On The Effectiveness Of Visible Watermarks” disclosed this vulnerability and ways to help the photography as stock image communities protect their copyrighted images. “However, a fact that has been overlooked so far is that watermarks are typically added in a consistent manner to many images. We show that this consistency can be used to invert the watermarking process — that is, estimate the watermark image and its opacity, and recover the original, watermark-free image underneath. This can all be done automatically, without any user intervention or prior information about the watermark, and by only observing watermarked image collections publicly available online”, the research stated.

The research further reveals that the team behind watermark removal algorithm trained the software with enough public examples for identifying the patterns. Then, through a ‘multi-image matting’ process, it separates the watermark’s components from the image. Google’s algorithm was then able to understand the elements of the watermark like its opacity, structure, shadow, etc, enabling it to be removed from any photo.

In order to fix this and create stronger copyright protections for images on the web, the research suggested adding random elements to the watermark. The changes made in watermarks should be such that it will leave visible artifacts after being removed from an image, including random geometric perturbations to the watermark, effectively wrapping the text and logos which are being used, suggested the team. The research also suggested that removing artifacts accurately is far more challenging.

Admitting that the defence isn’t perfect one, as there are always likely to be more sophisticated algorithms developed in future to bypass these practices, the team said that the current state of image protection is in a very sad state. Just a little bit of right kind of randomness can go a long way in keeping images safe on websites for a long time.