I am having trouble understanding how a vector for each pixel is more efficient then just an image. Both are vectors for each pixel in the image.

Is this in the context of motion only? So in that case there are t copies of the pixels and vector would be more efficient.