In fact, in the optical theory, watermark image can be perfectly restored, we do not talk about theory, only talk about operation.
Let's look at a watermark-added picture.
To get rid of the watermark, we want to determine two points, first, what is the original color of the watermark? Secondly, how much transparency is?
These two points are really a problem, a big problem, so I will write another article to explain. Here, let's guess, from the graph, the color of the watermark may be red (FF0000); To confirm this, we create a new layer, draw a red watermark to adjust the transparency, to see what the value of the time, and the image of the effect as near. BTW, when most people adjust their transparency, they like to use integers, like 20%,50%,85% ...
Did you see it? 60%, and the original watermark is exactly the same, you can enlarge your watermark and the original watermark overlap parts, with a straw to judge.
ff0000,60%. OK, OK, these two points, we are expected to restore.
Back to the original watermark picture, we select the watermark selection, create a new layer and fill the red. (Remember to save your selections for use)
Adjusts the order of the two layers, adjusting the layer mode of the original background layer to the difference value .
Select layer One, adjust the layer output color to 255x60%=153.
Flatten the layers, load the watermark selections, and adjust the output levels to 255x (1-60%) =102.
What did you see? The watermark is almost perfectly gone, yes, almost, although PS is not perfect theoretically, but enough to fool our eyes.
Look at the comparison:
If you want to understand the rationale for these operations, you can check out some basic optical books.