Google builds an algorithm to remove watermarks from images

watermarks


Google researchers developed an algorithm that completely removes watermarks from images. The algorithm works entirely automatically. According to Google, a vulnerability in the watermarks is used by stock imagery sites that makes it possible to remove the transparent stamp used to protect copyright.

However, the researchers identify a glaring error in this approach and exploit it to negate the visibility of watermarks completely. The method essentially takes advantage of watermarks by designing, apply to photos in the exact same manner.

Watermarks typically added in a consistent manner too many images. This consistency can use to invert the watermarking process, that is, estimate the watermark image and its opacity, and recover the original, watermark-free image underneath.

Google brings its Anti-phishing security to its Gmail app for iOS

algorithm


However, changing the position or opacity of a watermark does not impact on the algorithm’s ability, to remove the watermark from images. Researchers say, Randomization required for securing images.

While, manually rubbing watermarks from photos could take several minutes even with popular Adobe Photoshop. So, the researchers train a model to perform the same task much quicker and without any intervention from humans.

Google ‘PAIR’ improves AI and Machine learning

Google explains that the trick was to teach the computer how to accurately identify the repeating watermark structures and then fine-tune it how to replace them convincingly.

Researchers Tali Dekel and Michael Rubinstein, explain, if a similar watermark embedded in many images, the watermark becomes the signal in the collection and the images become the noise, and simple image operations can use to pull out a rough estimation of the watermark pattern.

However, these findings helpful for the photography and stock image communities.

[td_block_video_youtube playlist_title="algorithm to remove watermarks" playlist_yt="02Ywt87OpS4" playlist_auto_play="0"]

Comments