Even with the resource constraints of the average mobile device, you can often have the operation complete in less time than it would take to transfer the image, particularly under field / non-LAN network conditions.
One simple way around this would be to specialize in more intensive operations, particularly with parallel algorithms that can be done quickly by throwing 50 workers at it. Or alternatively, you could look for situations where the user needs to process a large batch of images asynchronously (then throw 50 workers at it).
There is a lot of room on the market for image processing tools - software, libraries, APIs, whatever. But basic operations which are already well-supported by many drop-in FOSS libaries are usually going to be hard to bring to market.
Add something extra which solves problems unique to the customer's problem domain or workflow. Then you become "the only" instead of "yet another."