It seems hard to say what they're really seeing there - deconvolution to the extent they're doing can be incredibly fickle. You need to know your camera nearly perfectly and have lots of SNR, to start. So it might be more likely that the highly unusual dark region not previously observed is actually a processing artifact.
Very interesting images nevertheless - we'll see how it bears out.
Deconvolution of astronomical images has been studied since the Hubble "astigmatism" incident. That was later corrected with more hardware "up there", but until then deconvolution was the only way to improve the quality of the images. From then on, the technique has been improved, thanks to better math and more processing power available in computers.
It isn't the same as getting closer, but it better resolves details of the image. I think their disclaimer about fine features is sort of a "we're saying this now so nobody screams ALIEN COVERUP at us later" (well maybe not that specifically, but there's always someone that sees one image and interprets it in his own way -- take Ceres Bright Spots for example).
Do you have a link to share for more info on what exactly they're doing? (or even some raw images next to some deconvolved ones?)
Regarding SNR, I believe these images are being sent lossless (whereas the preliminary high-res images we're getting in a few weeks will be lossy for now). Presumably they have a good profile of their camera from before launch too, knowing that they'd have to do this. So I wouldn't be surprised if this analysis does turn out to be accurate. Let's hope!
The 3rd image in the original article compares the original and the deconvolved result.
The main idea is you want to invert a linear transformation (a matrix) which is given by the operation of your imaging system, i.e. a blurring operation. This linear transform often has a poor condition number, however - this means that you can't really invert it, or if you try to you end up with certain components of your signal that need huge amplification, which makes noise sensitivity a major issue.
One solution to this is to introduce a "prior" on your signal that represents your expectations about what the signal looks like, i.e. is it smooth, or does it have a few edges etc. Then you can better tolerate the inherent sensitivities. I'm not sure what priors they're using for this situation or if they're using any at all, so in the absence of such a description I'm taking the results with a grain of salt.
Very interesting images nevertheless - we'll see how it bears out.