While such a thing might be hard to arrange in the case of a stolen phone, it might not be. How many people have their phones filled with images of themselves and with short video clips? Of course, the technology isn't even in the field yet, so it's impossible to say whether it could be fooled in such a way.
Fortunately, one of the other questions that are being raised in regards to this technology, concerning privacy, is probably not an issue. It's unlikely that MasterCard has plans to store actual images of each user in this program on their servers considering the amount of data involved. This is especially the case when the numeric representation of the image is what's really needed by the facial-recognition software, and not having to convert from a photo each time would dramatically reduce the workload and the related latency.
It's also important to remember that the idea of using a short video clip or other selfie as a means of identity verification is only a companion method. It's unlikely that MasterCard is planning to use images as a stand-along means of identity verification, if only because of the number of identical twins in the population. The idea of using a selfie will certainly appeal to a part of the population—but hopefully not to that part of the population that's made up of evil twins.