Google OCR cannot recognize punctuation marks out of line in Chinese vertical text.
For example, all periods (。) were missed in https://zh.wikisource.org/wiki/Page:CADAL09006598_%E9%80%9A%E4%BF%97%E6%96%B0%E5%B0%BA%E7%89%98.djvu/6 .
This punctuation layout is common in Chinese books published 100 years ago.
This is a problem of Google OCR and Google should be contacted to solve this.
You can keep an eye on the Vision API release notes, which mention when the Google OCR models are updated. And in fact, there might be an improvement coming in February, judging from this email they sent yesterday:
We’re writing to inform you that starting February 19, 2024, we are upgrading our Vision API Text Detection and Document Text Detection (OCR) models.
What do you need to know?
We are upgrading our Vision API Text Detection and Document Text Detection (OCR) models to Google’s latest OCR model, which improves text detection and recognition accuracy.
What do you need to do?
No action is required. Vision API OCR features will be automatically upgraded.
You can start using the upgraded models now by specifying "builtin/latest" in the model field of your Vision API requests. The current models will continue to be supported until February 18, 2024, after which the upgraded models will become the default.
If you need to continue using the current models, you can specify "builtin/legacy" in your requests. The legacy model will be deprecated after an additional 90 days.
We encourage you to test the upgraded models as soon as possible to ensure you are prepared for the upgrade. You can learn more about Vision API Text Detection, OCR and testing here.