what parameters can make the recognition faster ?

[mh cbon]

Hello,

I m inexperienced in OCR with tesseract.

i was thinking about writing a small application

that is able to recognize text over a video feed

taken from webcams (phone/laptop).

However i have tried online version (https://tesseract.projectnaptha.com/)

and its does not look fast enough for this specific usage.

I have read earlier that one of the factor of the speed is the size (px) of the input.

I wonder if there are more parameters that could change this behavior in a good way.

I had some ideas, but i m unsure that they are meaningful,

- use a specific trained model for one specific font, 

- a smaller alphabet,

- a smaller range of font sizes,

Could it help ? Are there other parameters i might investigate ?

I also wonder which has the most impact on the speed between recognition and preprocessing ?

As i want to read a video feed from webcam, i must assume preprocessing

will be required and i m afraid heavy.

However the example i have linked above does not seem to require much preprocessing.

thanks for your inputs.