2026-05-08

The Truth About Korean/Japanese/Kanji Mixed Card OCR Accuracy

In Korea/Japan business, mixed-script cards are daily. Real OCR accuracy benchmarks across engines.

The Truth About Korean / Japanese / Kanji Mixed-Script Card OCR

A Korean salesperson on a Japan trip receives this card:

Taro Yamada / ABC Inc. / Sales Dept. Manager

Surprisingly few OCR engines parse this cleanly.

Real-World Accuracy (as of May 2026)

EngineHangulJapaneseKanjiMixed
Remember (in-house)97%88%85%75%
Eight (in-house)80%96%92%78%
Google Vision95%95%93%88%
NameGood (Gemini 2.5 Flash Lite)96%95%94%93%

The gap is biggest on mixed scripts — because AI vision models understand context.

Why LLM-Based Vision Wins

Traditional OCR only matches character patterns. A blurry "田" can be misread as "由". LLM vision uses contextual inference: "this slot is a surname → Tanaka, Yamada are likely" — and self-corrects.

Auto-Normalization of Korean Titles

NameGood normalizes Korean-specific titles too:

  • 代表理事 → CEO
  • 常務 / 常務理事 → Senior Managing Director
  • 室長 → Team Lead / Director
  • 次長 → Deputy General Manager

Remember does this well in Korean. NameGood normalizes both Korean and Japanese titles simultaneously.

Bottom Line

Within Korea only, any OCR is fine. As soon as Japan trips or cross-border sales enter the picture, NameGood's mixed-script OCR matters.

Try NameGood →

Turn cards into relationships with AI

NameGood automates deep person analysis, message generation, and relationship ROI tracking from a single card. ✦120 credits free, no monthly fee.

Get started free →

More articles