Removing watermarks. Google has developed an algorithm to automatically remove watermarks. And protection from it

On this moment Photo watermarks are the most popular standard for copyright protection. They allow you to identify the real author and prevent you from using the image illegally. However, the effectiveness of watermarks should be questioned.

As Google showed in its study, removing classic watermarks is quite easy. Weak point is the similarity in the processing of each photograph. It is enough to figure out once exactly how the watermarks are located, and it becomes possible to create an inverse algorithm that gives the output an original image.

To remove watermarks, a large number of photographs with the same watermarks are run through neural network to identify similarities. Watermarks become a common feature of the sample, and the photographs themselves are identified as noise. After identifying the watermark, it is necessary to determine its intensity. Now the superimposed watermark can be subtracted from the image to obtain the original image.

Google offers a solution

Fortunately, Google not only identified the problem, but also suggested good decision. If you change the watermarks randomly each time, distorting them a little, then restoring the original becomes much more difficult.

It becomes obvious that even a minimal change in a watermark, such as distortion of the font, leads to this method into disrepair or make fraud much more difficult.

A watermark or mark - call it what you want - is a kind of signature of the author under his works. Some sites also watermark their images.

Often such inscriptions prevent us from using pictures downloaded from the Internet. I'm not talking about piracy now, it's immoral, but just for personal use, perhaps for making collages.

Removing an inscription from a picture in Photoshop can be quite difficult, but there is one universal method that works in most cases.

I have such a work with a signature (mine, of course).

Now let's try to remove this signature.

The method is very simple in itself, but sometimes, in order to achieve an acceptable result, additional steps must be taken.

So, we have opened the image, create a copy of the layer with the image by dragging it onto the icon shown in the screenshot.

Now it's time to analyze the inscription.

As you can see, the background under the inscription is not uniform; there is both pure black and various details of other colors.

Let's try to apply the technique in one pass.

Select the inscription as close to the text boundaries as possible.

Then click right click mouse inside the selection and select the item "Perform Fill".

In the window that opens, select from the drop-down list "Content-Based".

And press "OK".

Deselect ( CTRL+D) and we see the following:

There is damage to the image. If the background had been without sharp changes in color, even if not monochromatic, but with texture and artificially applied noise, then we would have been able to get rid of the signature in one pass. But in this case you will have to sweat a little.

We will remove the inscription in several passes.

Select a small section of the inscription.

We fill according to the content. We get something like this:

Use arrows to move the selection to the right.

Fill it again.

Move the selection again and fill it again.


Now select a tool "Brush" with hard edges.


Press and hold the key ALT and click on the black background next to the inscription. We paint over the rest of the text with this color.

As you can see, there are remains of a signature on the hood.

We will paint them over with a tool "Stamp". Size is adjustable square brackets on keyboard. It should be such that a piece of texture fits into the stamp area.

Clamp ALT and click to take a texture sample from the image, and then move it to the right place and click again. This way you can even restore a damaged texture.

“Why didn’t we do this right away?” - you ask. “For educational purposes,” I will answer.

We have discussed perhaps the most difficult example of how to remove text from a picture in Photoshop. Once you master this technique, you can easily remove unnecessary elements such as logos, text, (garbage?), etc.

For those who often work with video, a pressing issue is removing a watermark or any inscription from a video. These can be videos that contain TV channel logos, a special sign for a YouTube channel, or just text. To remove a watermark from a video, use special programs which I will tell you about today.

Remove watermark from video

A watermark is most often a translucent small image that is located in the corners of frames, but can also be in other places. Such signs indicate the involvement of the video in a particular organization and may have a simple informational meaning. A sign often serves to recognize a brand so that viewers can identify the author by it. It is also worth considering that in some situations, removing watermarks is prohibited by copyright law. But there are also technical signs that indicate the beginning of the recording, for example, “REC” or a display of the time period in which the video was filmed.

Removing the watermark using Remove Logo Now

Remove Logo Now can cut out any static text on a video frame. One of its strongest advantages is its ease of use. Everything in it is extremely clear; the toolbar has only a few buttons that are necessary when working with the program. You don't need to spend a long time getting to know professional video editors because there is a great tool that was created to remove watermarks from videos.

The program was created based on the Inpainting algorithm. This technology allows you to remove subtitles, TV channel logos, video date and time, and any other static objects. Even in the most difficult situations, only invisible artifacts may remain. In order to use this program, go to the official website of the editor https://www.removelogo.com/ru/remove-logo-from-video/ and select the required version. Download and install the program.


Erase video logo using Avidemux

Avidemux is a program that also allows you to erase logos and remove watermarks from your video. This is a fairly simple and easy-to-use editor that has useful features. In addition to removing watermarks, the software can:

  • add visual effects and various filters to the video;
  • convert videos to other formats;
  • export and import audio streams to video;
  • tool for working with subtitles;
  • work with the command line.

Avidemux is available on the website http://biblprog.org.ua/ru/avidemux/. To use the program:

  1. Install and run it on your computer.
  2. Drag the video from a folder on your computer into the working window of the video editor into the black square.

      Drag the video onto the Avidemux working window
  3. In the “Video Output” block, select “Copy” and select “Mpeg4 ASP (xvid4)” from the drop-down list.

      Select video output from the toolbar
  4. Now go to the “Filters” item. It is located just below the “Video Output” block. In the window that appears, select “Sharpness” in the left block, and “Mplayer delogo2” in the right block, double-click on it.

      Selecting video sharpness
  5. Next, by setting the values ​​in the upper left corner or simply selecting with the mouse, set the area where your watermark is located.

      Specify the coordinates of the area where the watermark is located
  6. When you decide on the area, pay attention at the bottom of the window there is a “Preview” checkbox; check it and your watermark will disappear.
  7. Then click "Ok" to confirm your actions. This way you can remove as many watermarks, signatures, logos and more from any video.
  8. When the video is ready and all watermarks have been removed from the video, click the floppy disk icon at the top to save the video.

So, lately a lot of pictures on the Internet are protected with a watermark (aka Watermark). It is not clear why they do this, because there is almost no sense in it. A watermark does not give you copyright on the image, and if it contains a link, less than 0.1% of people will still enter it.

The reasons for removing Watermark can be different, and now I will tell you how to remove a watermark from a photo quickly and without problems. We only need one program, 8-10 megabytes in size, called Inpaint. At the time of writing (see date) of this article, latest version 6.2.

Inpaint interface and functionality

The program is simple, there is nothing complicated in the interface. Inpaint is actually designed to remove watermarks from images and photographs, but if you wish, you can cut out any object.

All we need to do is to select with a marker or other tool the area from which we will remove unnecessary elements, after which we use the same marker to select the donor, that is, the area from which the data will be selected to be filled.

It is best, of course, to remove them where the background is uniform, but if the watermark is located right in front of a person’s eyes, then complications can arise.

Actually, watch this video, here I will show you how to remove a watermark from any picture. I myself applied two watermarks to the image, one in the form transparent pictures, the other in the form of an inscription. It turned out a little crooked, but overall Inpaint coped with the task.




Example of how the algorithm works

Tali Dekel / YouTube

Researchers from Google Research have developed an algorithm to automatically remove watermarks from stock images. Since such images typically have the same watermark, an algorithm based on many such images can calculate its outline with high accuracy and then cut it out without leaving a trace. At the same time, the researchers proposed a defense against such illegal acquisition of images - to create a unique watermark for each image with slight differences. The study is reported on the company's blog.

Many sites, including the media, do not hire their own photographers, but use the services of photo banks - special sites that collect millions of photographs on a variety of topics. To allow the user to evaluate the photo before purchasing, many of these sites show the photo in reduced resolution with a watermark applied. It doesn't interfere much when selecting an image, but it can be difficult to remove due to its translucency.

Researchers from Google showed fundamental vulnerability most photo banks. The fact is that usually a photo stock site uses the same watermark for all images. Since watermarks often consist of many symbols and lines, it is quite problematic to determine it qualitatively from one image. Developers at Google solved this problem by feeding the algorithm a lot of images. It found matches in images and turned them into a signal and the rest of the image into noise. Thus, based on hundreds of images, the algorithm obtained the exact contours of the watermark and its transparency. Having this data remove the watermark in automatic mode no problem.


Researchers have demonstrated the successful operation of the algorithm on popular photo banks such as Adobe Stock or Fotolia. But the goal of the work was to create protection against this method of obtaining images. To do this, the researchers tried several options. Randomly moving the watermark or its transparency had virtually no effect on the algorithm's performance. However, it turned out that even an almost invisible random distortion of the contours of a watermark, for example the contours of fonts, causes the appearance of many noticeable defects in the final image. The more changes made, the more difficult it became for the algorithm to remove the watermark. The researchers hope that their work will help photo bank owners combat piracy more effectively.

Google researchers are not the first to propose using randomness to protect against piracy or counterfeiting. For example, British scientists decided to protect documents random distribution cellulose fibers in paper into a digital key unique to every sheet in the world. And Korean scientists use random processes at the micro level to create unique and unforgeable identifiers.

Grigory Kopiev



2024 wisemotors.ru. How it works. Iron. Mining. Cryptocurrency.