I built Tesseract3.03 from source on an Ubuntu 14.04.2 machine. There is a page describing how it can be training on new languages/fonts, but what I can't figure out if Tesseract can be trained incrementally. I have some test images that come with accompanying ground truth (UTF8 text files that contain the text shown in the images). I ran Tesseract on these images using the out of the box model for the language (English) that came with the source distribution. Some were recognized well, some badly. So what I want to do is to take some of these bad images and use them with their ground truth + the original English model to make a better English model that would do better on the bad images. Can I do it? Thx!