There are many reasons why stock photo distributors would be squirming in discomfort. Google has come up with the new algorithm that completely erases watermarks. It actually becomes pointless to use watermarks for preventing the access to clean images. Sad but true, the Google update gives way to unauthorized use no matter who the person might be.
While professionals insert watermarks for protecting copyrights, the researchers revealed loopholes. This permitted them to remove marks automatically. It finally led to a way of recovering images with higher accuracy and clarity.
Method used to remove watermarks
Much before everything is carried out, professionals determine the image structures that belong to a common watermark. This might seem simple but it’s actually a tedious task. Numerous detection steps are involved in identifying the watermarked regions. Specifically, the estimation as well as the detection steps are executed again and again.
- A consistent image gradient across the collection is initially determined. Later, at every pixel location, the median of the gradients is calculated along the vertical and horizontal axis. With the increase in the number of images, the median converges to true matted watermark.
- With the calculated median, the watermark is detected using Chamfer Distance. Later, the verbose edge map is obtained for the respective watermarked image. Furthermore, the watermark position is considered to be the pixel with minimum distance on the map. This detection method served to be beneficial because it offers detects the marks quickly even when there’s a change in the opacity level.
- Once the aligned detections are known, experts then move on in finding a solution for multi-image matting problem. In this situation, the intensities are decomposed into watermark, alpha-matte and original image components. It becomes essential to accurately estimate every component because small errors can lead to the visibility of watermarks.
- Finally, after fetching the required solutions, Google moves ahead in removing the watermark. Just in case the desired solution is not obtained, then every step is carried out all over again. Rather than opting for direct reconstruction, image-watermark decomposition is always preferred.
Everything that’s mentioned above has been included in ‘On the Effectiveness of Visible Watermarks’, a paper presented at 2017 Computer Vision and Pattern Recognition Conference. While this Google algorithm helped stock photography markets to enhance brand image, the Google team used the trend to develop an algorithm to edit out signature marks.
The end result was different and something worth noting. The researchers performed the task in no time without the interference of human beings. It also saved time else it might have taken minutes to wipe off watermarks through software like Photoshop.
Once the trick was devised, the professionals at Google trained the computer to identify the watermark structures with accuracy. They then elaborated on how to replace them. A rule was followed according to Tali Dekel and Michael Rubinstein. If there is a similar kind of watermark with many images, then the mark became the signal while the images were imagined to be ‘noise’. Image operations helped to give a rough idea of the watermark pattern. On the other hand, stock vendors were fortunate because researchers also came up with an effective way to protect photos. Though they succeeded in removing marks, slight modifications made it tough for the algorithm to hide signature marks.