It basically runs SWT on the image, and creates a 3d Lab histogram of the colors the SWT marked as text. Then it does a morphological dilation of 10 pixels and subtracts the original mask to get the colors of the pixels that represent the background.

Then it just binarizes the image by whether the internal histogram is larger than the corresponding value of the color on the external histogram.

It's a strategy that works quite well on machine-printed text, but probably less effective than existing strategies when it comes to scans or photographs.

Curious about this too. Also, what's the stack providing Tesseract-as-a-Service? According to my cursory search, Google app engine won't run Tesseract as its a native library, not an API. I'd like to try this on non-Latin/CJK hardcoded subtitles, but ocrat does latin only.

Thanks, would also appreciate recommendations for better/faster Tesseract results. Particularly with scanned document PDFs.