When written in Japanese characters, addresses start with the largest geographical entity and proceed to the most specific one. When written in Latin characters, addresses follow the convention used by most Western addresses and start with the smallest geographic entity (typically a house number) and proceed to the largest. Big and little-endian, in other words.
I don't quite get why the latter became widespread in the west, though - we read text from top to bottom, we pinpoint addresses by drilling going from the largest geographical entries down to the smallest, why make them go reverse against each other?
One benefit I can think of is that the volume of addresses that the average person reads tends to decrease with distance from the person's home. This means that in a left-to-right, top-to-bottom language, you can start reading from the beginning of the address, and as it gets more general, eventually you hit a point where it becomes identical to your address for the rest of it, so you can stop reading. For example, when reading an address in the city I live in, I can read the number and the street, and then once I see the city name, I can stop, since I know everything I need to. This would be a bit more difficult if you had to start from the bottom right and work your way backwards, so I'm guessing that the ordering was chosen to maximize efficiency for those common, local cases.