I was watching the feature film Ten Canoes on DVD with a friend. It's a fantastic movie, very unusual. It won a prize at the prestigious Cannes Festival. However, the picture quality of the DVD leaves something to be desired. I can't be absolutely sure, but I think it was encoded on a PC.
The problem in the encoding is pretty simple. The motion vector search range used during the encoding was set to a window of 16 pixels. In MegaPEG, you'd get a search window this size if you slide the Motion slider all the way to the top (very low motion).
The problem is, that although this feature does not have a lot of action in it, it does have motion. Camera motion, closeups with peoples heads moving through the frame. That kind of thing.
Here are three screen shots from the MPressionist analysis of the title. The first two show the two pages of motion vectors. Blue and green represent motion vectors that are pointed forward and backward in time. As you can see, no motion vector exceeds the 16 pixel maximum used during the encoding.
The encoder used on this job, handcuffed by the small motion vector search window, had to resort to other techniques to get the video into the required bitrate. Encoder theory predicts that in order to get higher motion passages to encode, more compression will be required. In this case, extreme amounts of compression (indicated by peaks in the third picture).
I have no doubt the encoder used on this job was capable of getting the picture quality right. Although the content is highly textured, they used an 8.6 Mbits/s bitrate ceiling, plenty of bits for something like this. Instead, the difficulties here are the result of "not knowing" what the encoder did. Without that crucial feedback, it's easy to miss the artifacts in this title, which appear only during high motion scenes and do not quite get blocky. Instead, high motion areas end up kind of a blurry mess.
Don't take my word for it. Rent it and watch it.
