This blog has discussed several options for performing OCR on Chinese texts, but the options all required a desktop or laptop computer (Google Docs, Adobe Acrobat, Sciweavers i2OCR). In this post, we’ll look at several options for OCRing Chinese on iOS devices.
The contenders:
- ABBYY TextGrabber + Translator is an OCR app that supports many languages and ties into Google Translate.
- LRDict is a Chinese dictionary app that has an OCR feature. I used the lite version for testing.
- Pleco Chinese Dictionary is an app with several tools for studying Chinese, primarily a free dictionary; OCR is a paid add-on.
All of these require starting from an image file, not a PDF. You can take a photo with your device’s camera or import one from the Photos app. My tests used two different photos taken with an iPhone 4S and iPad 3, so your results may vary.
Round one
A page from an adaptation of 西游记 Journey to the West. This book has a rounded font style that I thought might be tough to OCR.
A bit of the text from Round 1 showing the font
ABBYY TextGrabber: TextGrabber made about 13 character recognition errors in about 200 characters. Given its misreading of 中 zhōng as a circled 1 ①, the roundness of the font may have been problematic.
LRDict: LRDict had even more problems with this text, repeatedly misreading common words like 的 de, 有 yǒu, 很 hěn, 和 hé, 天 tiān, 各 gè, 开 kāi, 大 dà, 在 zài, and 日 rì, among others. OCR is seldom perfect, but this result was unacceptable, with over a third of the characters wrong.
Pleco: Pleco’s errors were almost all with punctuation. It consistently failed to recognize the Chinese period ( 。) and serial comma ( 、), as well as quotation marks. Otherwise it split 很 into two characters (彳and 艮) several times (although it recognized 很 successfully elsewhere), but made no other errors with actual hanzi.
Round one winner: Pleco
Round two
A page from 第三只眼睛 The Third Eye, the first title in level 3 of the Chinese Breeze series of graded readers (more about the series in this post). This text was tricky because it includes footnote numbers, but if you’re scanning academic texts, this might be a real issue.
ABBYY TextGrabber: TextGrabber did much better in this round. Only one hanzi was misread (perhaps because it was at the edge of the page) and all the punctuation was correct. Otherwise TextGrabber only had a problem with the number 40, which was read as a 4 and something else (either a D or a Chinese period).
LRDict: LRDict also did better, but was still the worst of the three. Five hanzi were misread in the majority of the passage, but the last couple of lines got increasingly garbled. There were also problems with the numbers and punctuation.
Pleco: Pleco misread four hanzi, but three of them were 是 shì (the fourth error was giving 入 rù for 人 rén). Periods were not a problem this time, but quotation marks were still always wrong. The number 40 was misidentified as various Chinese characters, although 2 and 39 were fine.
Round two winner: ABBYY TextGrabber
Tiebreaker round
Catullus 13. This began life as a PDF scanned (on a Ricoh Aficio MP 5000) from a Chinese edition of the poems of the Roman poet Catullus. The PDF was then saved as jpeg image (remember that these apps cannot OCR PDFs directly). The quality of this image is clearly worse than that of the pictures taken with the latest iPhone/iPad cameras.
ABBYY TextGrabber: 12 hanzi were misread and a string of three was missed entirely.
LRDict got disqualified from this last round because after OCRing, I was unable to select the whole text for copying/pasting into another app. It was out of the running already anyway.
Pleco: Once again, Pleco had some problems with punctuation, but it only missed one hanzi; it didn’t see an instance of 一 yī.
The overall winner: Pleco!
Just considering the accuracy of recognizing Chinese characters, Pleco was the most reliable. If the developers work on its handling of punctuation and numerals, there would almost be no contest.
The ultimate decision as to which app to choose may well depend on why you are scanning the text. Do you want to have it machine-translated? Do you you want to read it with the help of a dictionary? Do you want to extract words and add them to a flashcard app? With the final goal in mind, you may want to consider the other features of the apps in question.
ABBYY Textgrabber offers translation of the whole text into a number of languages (powered by Google Translate). An action button gives you one-tap options to copy, e-mail, post to Twitter and Facebook, send to Evernote, and more.
LRDict was noticeably slower than the others. A bigger problem was the fact that the “move and scale” options did not allow me to get the entire photo into the app’s squarish OCR area (why not match the dimensions of an iPhone photo?) Even the last lines that fit in the OCR area came out very garbled. And, as mentioned above, I could not always access all of the OCRed text.
If these problems somehow don’t affect you, LRDict has some features that could be useful, if they worked properly. Tapping on the image (not the OCRed text) brings up a version of the text that is color-coded by tone and has ruby pinyin. Here you can tap a character to look it up in the LRDict dictionaries (cc-cedict for English). For some reason, however, this view would only display the first part of a text.
Pleco’s OCR offers benefits by integrating with other features of the app, namely its dictionaries and flashcards. These will be discussed further in an upcoming post.
