Translating from, I interpreted your issue. Perhaps you could try the dispatcher/preformer model with state machines.
The dispatcher uses imageFound activity for each image you need to find, set the output Found variable and write to queue. The performer will then pick up from the queue and process it.