Re: [Bug-ocrad] Errors in 114x26 (wxh) images containing 5-6 uppercase letters w/ ocrad v0.20

[Antonio Diaz Diaz]

Hello Tom,

Tom Littauer wrote:
> My wife likes to play the Jumble (tm) word scramble game in the daily 
> newspaper, but wants the answers the same day (they're usually reported the 
> following day).
>
> I built a little script to extract the word images, feed them individually 
> through OCR and then through an anagrammer.
>
> I started by using ocrad (v 0.20 in OpenSuse 11.4) but noticed some failures. 
> gocr works just fine, especially as I can restrict the character set to A-Z.
>
> I can fairly easily modify the script to send you images everytime gocr and 
> ocrad disagree (samples attached). So far the rate is 2-3 failures per week 
> out of 24 images per week.
>
> Would this be useful to you?

Of course, thanks. It is always useful to have samples of misrecognized 
text. Specially simple ones like yours.

But, please, send them to my private address instead of to the list.

Meanwhile, you may find that giving the --scale=2 option to ocrad 
improves the results. (The letters in the samples are a little small for 
ocrad).


Best regards,
Antonio.