What’s a good way to throw off duplicate image identification bot?
 

Crop, apply some filter, add few pixels here and there, …?

If we're talking about like a spyder that goes through webpages, in some cases it would probably be enough to just load the image in via javascript asyncronously - so
<img src='somegenericimage.jpg' data-src='theactualimage.jpg'>

and then load it in with query from data-src.

You talking about Google reverse image?

Crop the image (not a resize based on percentage but a specific area cropped), lower the quality, apply an Instagram filter, etc.
WG