Basic Information

Tesseract 5.3.0

Operating System

No response

Other Operating System

Windows 7

uname -a

MINGW32_NT-6.1-7601 PC 3.3.6-341.x86_64 2022-11-20 15:12 UTC x86_64 Msys

Compiler

GCC 12.2.0

Virtualization / Containers

No response

CPU

Intel Core i7 Q720

Current Behavior

If tesseact is built without the legacy engine (--disable-legacy), recognizing vertical Japanese text with jpn_vert (from tessdata_fast) and PSM_AUTO (--psm 3) gives garbage. Here is an example image:

Here is the output of tesseract 1.png stdout -l jpn_vert --psm 3:

…4

/
\

09$2pY
コ べ
ほり メー14

Here is the correct (albeit with some OCR errors) output that I get either without --disable-legacy, or when using PSM_SINGLE_BLOCK_VERT_TEXT (--psm 5) explicitly regardless of whether the legacy engine is enabled:

ケー タイ は
カバ パ バン に
入れ て ある し

Expected Behavior

No response

Suggested Fix

If this is a kind of feature rather than a bug, it should probably be documented somewhere.

Other Information

No response