Tue, 02 Dec 08

OCR on the Mac with Tesseract

I wanted to OCR another letter this morning and, as before, I used tesseract for the job.

I ended up spending some time testing the quality of the output (the generated text) against different quality inputs (as defined by the dpi setting in my scanner software). The results were not surprising. The best output, with only a couple of errors in the whole letter, came from the 300dpi scan (the highest I tried). The scan at 150dpi contained a handful of mistakes and at 96dpi the output was, in places, complete gibberish.

In case this is of use to anyone I have copies of the, smaller, second page of the letter I was scanning, along with the OCRd output. The scanner is an Epson Stylus DX4800 and I used seashore to convert the images, from colour jpegs, to greyscale tiffs.

At 300dpi

Nothing is ruled out in terms of further development, and I have met with a representative of mySociety to discuss ways of continuing to improve what we do.
I am fully supportive of the aims of the Free Our Bills campaign, and I am keen to work with the relevant House authorities to see what can be done to open up the legislative process to a broader audience.
I hope that this is helpful to your constituent.

At 150dpi

Nothing is ruled out in terms of further development, and I have met with a representative of mySociety to discuss ways of continuing to improve what we do.
I am fully supportive ofthe aims ofthe Free Our Bills campaign, and I am keen to work with the relevant House authorities to see what can be done to open up the legislative process to a broader audience.
I hope that this is helpful to your constituent.

At 96dpi

Nothing is ruled out in tcmts cI` further develcpntcnt, and I have met with u rcpresentmivc ul’ mysociety tn discuss ways nfccnvsinuing to improve what we dc.
I am titlly supportive oI’thc uirns ofthe I-`rce Our Bills campaign, und I urn keen vn work with the relevant House authorities to sec what can he done to open up thc legislative process In at broader audience.
I hope that this is hclphtl to your constituent.