There is a classic channel-capacity notion that exist based on the work of Shannon that gives gives measure of amount of information that can be passed through the channel given its properties.
In the context of watermarking, the image or signal where information is hiding can be thought of as channel. While, in traditional Shannon model the channel distorts the signal if the bandwidth of channel is constrained. Here, if we keep adding more information of the watermark the channel (the host signal) gets distorted.
The question is, is there a notion of what should be the limit or capacity of information (bits) that can be added under a given host image?