Although the traditional image watermarking method is efficient, it damages the details seriously. De-watermarking is easier said than done. Some watermarks can be repaired in a few seconds, and some watermarks may not be repaired in an hour or two.

Some pictures that are not very rich in details can be filled with adjacent pixels by image processing software such as photoshop to cover up the watermark, which can achieve almost perfect results.

Facing some extremely complicated images, ps is no longer perfect.

Facing the complicated details of watermarking, the traditional ps watermarking method can no longer meet the demand.

Now with AI technology, removing watermark can be almost perfect.

With the continuous development of artificial intelligence technology, deep learning is more and more widely used in the field of image processing. At ICML20 18, researchers from NVIDIA and Massachusetts Institute of Technology demonstrated an image drying technology, Noise2Noise, which can automatically remove the watermark, blur and other noises in the image, almost perfectly recover, and the rendering time is milliseconds.

Paper Noise 2 Noise: Learning Image Restoration without Clean Data

The third-party reproduction project: yu4u/noise2noise, which can be used to remove subtitles and image noise, but the author did not add the watermark removal function.

I modified this python script, and it can be watermarked.