I had this theory that if you were to convert each frame of the bridge guy video to 32 bit, then align, reduce opacity, and layer these in Photoshop it would give you a clearer image of the suspect. And, it worked.
The video stills create so many artifacts that we see hair, a mustache, sagging jowls, a bulbous nose. But, when the layers are lined up, the "real pixels" are magnified while the false pixels are canceled out. The reason for converting the stills to 32 bits is to allow the artifacts to essentially get lost in the other color channels.
The biggest mistake of this investigation was releasing those stills. They create a false concept of what this perpetrator actually looks like. In reality, he looks much closer to the more recent sketch. I know we can't release modified versions of the images so I'm not doing that now. But, I will be contacting ISP next week to see if they can duplicate this independently.