I am currently working on a project in which i'm supposed to classify whether an image contains a translucent watermark or not. This is hard to do with standard object classification or template matching because of the translucency and very low image quality.
I thought about generating a dataset with the watermark pasted somewhere on the image. Given enough generated data, could I train a neural network to classify the real images? If not, why not? Thank you for your help.