Hi, I am working on tesseract. My goal is to make a real-time system. So I do want it to run on almost 20+ FPS.
I am getting almost 0.5s - 1.5 s on a single frame via tesseract. I have tried alot of things to improve this speed but of no use. Some of the tried things are
* fast integer models given by Tesseract
* tessedit_do_invert=0 argument
* gray-scaling (dont need further pre-processing as images do not have noise etc.
* 4 threads via OMP_THREAD_LIMIT
* 4 processes via tessedit_parallelize
But none of these gave any improvment except tessedit_do_invert=0 and gray-scaling. I do not want to loose any chracter so can not downsize it more.
I have implemented custom multi-processing to run 4 tesseract instances in parallel and it has boosted the speed a bit. I would need suggestions and techniques in-order to improve the latency of the OCR(cant skip the frames). I want to make it to at-max 0.1 seconds per image. If you do have any other OCR engine in mind, let me know. Thanks.