Used DevonThink3 to OCR a simple image – a screenshot of an english, typed powerpoint slide – to test its OCR capability to find that something isn’t going quite right.
The screenshot (png format):
Although the result (pdf + overlayed OCR’ed text) appeared superficially fine, the actual text generated wasn’t; it was gibberish. This gibberish can be more thoroughly observed by changing the OCR’s output format to “To Word Document” or “To Web Archive”. The generated web archive’s (html format) contains gibberish text like the following all the way through:
1^16 0001*56, 776^6 831(1 1^110^5
I “15 ^0^9 ^35 1116 83016 81^001:0^ 28 1^131 甘「011|3・‘‘
I “15 ^0^9 传 1500101*^^10 10 1^131
1-10^676^，^6*76 06^61^ 1*6311^ 5^6116：！ 001 1^16 ^613||5 360111 ^^131 1^115 016305,
\/\/& 77111 510：17 3 5^60131 IX？6 0^ ^061100 661^060 甘「。口|3& 03116：！ 2 110^10^10^1115^1，/\0
150^10^^115^1 18 2 8^60131 ^!36^0010010^^^1501，7^16 6^66^「00切'十0010” 3^6
“010「|3^” 10561^6^ 01630 4 53016 5^13^6/,
7^161*6 31^6 1^/0 511：031：1008 ^^161^6 ^0010010^^^15018 31*156：
・ 77^160 006 &「011？ 18 2 51165^011^ 0^ 3001^161^；
・ 77^160 006 &「011？ 18 2 ^0011601 0^ 3001^16
7^16 001*^8^00(^105 ^0010010^(3^15015 3佗 03116：！ 6016611：1|055 (^11011601： 012^)5-
7X180 10 1^118 0^13|31：6^，776 \^/||1 0001|3161：617《2530 ^101116 266(130 ^0^8, 300)3
Setting the OCR process’ output format to Word document also results in similar gibberish.
Would gladly pay for DT3 but at the moment something appears gravely wrong with the OCR.