Vertical Japanese Business Cards (縦書き): Why Scanners Break on Them

Hand someone a stack of Japanese cards to digitise and the vertical ones are where the batch goes wrong. Not dramatically — nothing errors out. The record simply comes back with a name that reads as nonsense, or a phone number one digit short, or a perfectly clean English contact with no trace of the Japanese that was printed on the other side.

This is the deep dive on that specific failure. For the general meishi workflow — readings, name order, follow-up etiquette — see the guide on scanning Japanese business cards. What follows is about layout: what vertical setting does to a scanner, and how to capture a 縦書き card so the result is worth keeping.

What is 縦書き, and why is it still on business cards?

縦書き (tategaki) is Japanese set vertically. Characters run from the top of a column to the bottom, and the columns progress from the right edge of the card toward the left. The first thing you read on a vertical meishi is the top of the rightmost column; the last is the bottom of the leftmost. 横書き (yokogaki) — horizontal, left to right — is what most modern cards use, and what essentially every OCR engine was built around.

Vertical setting survives because it carries a register horizontal setting does not: it is the layout of calligraphy, of formal correspondence, of most printed literature. Cards from law offices, construction firms and 工務店, traditional crafts and food producers, ryokan and other hospitality businesses, temples, and long-established family companies still arrive vertically set, often on heavier stock with a deliberately formal typeface. It also turns up on the Japanese face of bilingual cards whose English face is ordinary horizontal Latin — the two sides follow different conventions on purpose.

A common arrangement puts the company name, department and title in the right-hand columns, the personal name to their left in larger type, and the address and contact details down the left side or across the bottom. That last block is often set horizontally even on an otherwise vertical card, and it matters more than it sounds.

Why does vertical text break most business-card scanners?

Nearly every OCR pipeline starts from one assumption: text is a series of horizontal lines, read left to right, stacked top to bottom. Line detection, character grouping, reading order and the layout model that decides this block is a name, this block is an address all rest on it. A vertical card breaks the assumption at the first step, and everything downstream inherits the error.

What you see in the record	What actually happened
Characters in the wrong order	The column was read as if it were a horizontal line, or the columns were traversed left-to-right instead of right-to-left
Name and address fused into one field	Two adjacent vertical columns were treated as a single text block
One column split across two fields	Line-breaking logic written for horizontal wrapping cut the column in half
Kanji returned as gibberish	The image was rotated ninety degrees to force a horizontal read, leaving every glyph on its side
No Japanese at all — only the English side	The pipeline returned its highest-confidence reading, and the Latin face scored better

The last row is the expensive one, because it does not look like a failure. You get a tidy contact with a name, a company and an email, and no sign that the title, the legal company form and the authoritative spelling of the person’s name were on a face the tool ignored. The English side of a bilingual card is a company-chosen rendering rather than a translation, and it routinely flatters the title — the CJK titles reference covers how far apart the two sides drift.

What happens when one card mixes vertical and horizontal text?

Most real vertical cards are mixed, not purely vertical. There are two usual patterns:

• Across the two faces. Vertical Japanese on the front, horizontal rōmaji on the back. This is the standard bilingual meishi.
• Within one face. The name and title set vertically, the address and contact block set horizontally underneath or beside them. Designers do this because email addresses, URLs and international dialling codes are Latin strings that will not set vertically without being rotated or broken into pieces.

Both patterns punish an engine that picks one dominant orientation per image: whichever orientation loses the vote is read as though it were the other. And when the two faces are processed as unrelated images, you get two half-contacts — one holding 山田太郎 and 株式会社サクラ商事 with no email, the other holding “Taro Yamada” with no title. Merging front and back into a single contact is not a convenience on Japanese cards. It is a correctness requirement.

Which vertical typography details go wrong most often?

Vertical setting is not horizontal setting turned on its side. Several characters change shape or position, and each change is a place where a horizontal-first engine can slip.

Element	Horizontal setting	Vertical setting	Where it goes wrong
。and 、	Sit at the bottom-left of the character cell	Sit at the top-right	Punctuation attaches to the wrong neighbour, or is read as a stray mark
Small kana っ ゃ ゅ ょ	Bottom-left of the em box	Top-right of the em box	Read as full-size や or つ, which changes the word
Long vowel mark ー	A horizontal stroke	A vertical stroke	Competes with the kanji 一, with hyphens and with dashes; frequently dropped
Brackets 「」（）	Upright	Rotated ninety degrees	Read as different punctuation, or discarded
Numbers	Arabic digits in a line	Kanji numerals, stacked digits, or a 縦中横 run	Digits reordered, merged, or transcribed as kanji into a numeric field

縦中横 (tate-chū-yoko, “horizontal within vertical”) is the one worth knowing by name. It sets a short run of characters — most often a two-digit number — upright and side by side inside a single vertical column, so the number reads normally instead of stacking one digit per line. It is used for years, floor numbers, the 丁目 and 番地 components of addresses, and pieces of phone numbers and postcodes.

For a human reader it is an elegant solution. For an engine tracking a column as though it were a line, it is a small horizontal island in a vertical stream, and the digits inside it get reordered, merged with the numbers above and below, or lost. That is the mechanism behind most mangled Japanese phone numbers, and why the numeric fields deserve a specific check rather than a general glance.

How do I photograph a vertical meishi so it reads correctly?

1. Hold the card the way it was printed. Square-on, upright, top edge at the top. Resist the instinct to rotate it so the Japanese “reads like English” — rotation does not convert vertical text into horizontal text, it just tips every kanji onto its side and hands the engine a harder problem than the one it started with.

2. Light it evenly and kill the glare. Vertical cards skew traditional in production: heavier stock, letterpress impression, foil, embossing. Angle the card slightly away from the light rather than shooting head-on under a ceiling lamp; diffuse daylight near a window is the easiest good light there is.

3. Fill the frame and keep the columns straight. Column separation is what lets a scanner keep the name apart from the address, so give it clean geometry: close enough that the smallest characters — usually the address block — resolve sharply, with the card’s edges parallel to the frame so the columns stay vertical instead of converging.

4. Capture both sides. Two shots of the same card, merged into one contact. On a bilingual meishi the vertical face holds the name, title and legal company form while the horizontal face holds the email and URL; you need both, in one record.

5. Check the number fields before you file it. Compare the phone number and postcode against the photo, digit by digit, while the card is still in your hand. Five seconds, and it catches what 縦中横 produces.

What should I check after extraction?

A short pass over each vertical card, in order of how much a mistake costs:

• Name order and split. Japanese names are family-name-first: 山田太郎 is family 山田, given 太郎. Confirm the two halves landed in the right fields rather than one field holding all four characters.
• Phone digits, counted. Not skimmed — counted. A Tokyo landline (03-1234-5678) is ten digits; a mobile (090-1234-5678) is eleven. A number one digit short will not connect.
• Postcode. Japanese postcodes are seven digits, written 〒123-4567. Check both halves.
• Company legal form and its position. 株式会社 may be printed before the name (株式会社サクラ商事) or after it (サクラ商事株式会社). Both are legally identical, but the position is part of the registered name — it belongs where it was printed.
• The title. A title read out of a mis-ordered column can turn into a different rank entirely.
• Anything on the photo with no home in the record. Look for a second phone number, a branch line, a 支店 address — the fields a merged-column error tends to swallow.

Where BizCardPro.AI fits

BizCardPro.AI keeps the native script exactly as it was printed — vertical or horizontal — and puts a romanized reading beside it rather than replacing it, so the original characters survive the scan. Front and back are merged into a single contact instead of two half-records, the name is split into family and given fields on CJK ordering, and the photo of every face stays attached to the contact. That last part is the honest answer to vertical layout: it is a hard problem, nobody has solved it, and what makes an error fixable is having the printed card sitting next to the field you are checking.

---

Vertical meishi are a small share of any stack and a large share of its errors. Treat them as their own class — shoot them upright, capture both faces, give the phone number and postcode a deliberate look — and the rest of the batch takes care of itself.