Training tesseract for images having printed and handwritten data

[akhil katpally]

I am really confused with the tesseract training

I have used tesseract with eng language the results are 80% accurate, i am trying to improve the accuracy. I thought of creating a new language with orginal images as train data. Will the results improve if i use orginal data as the train data? I have trained it but the results are very bad when compared with the original.

So is there any way i can modify the orginal eng trainedata so that i can improve the accuracy? I tried following the docs,like changing the uncharambigs but it is of no use.

And another question is my image has both printed data(majority) and handwritten data. So, how should i train the tesseract, should i do it for printed data first and then for the handwritten data or do it seperately?