rohitsaluja22 / openocrcorrect Goto Github PK

How are we getting the output from Primary OCR and Secondary OCR? What are the Primary and Secondary OCR used (like Tessaract, etc.)?

Terminal log

dhaval@dhaval-Aspire-5750:/var/www/html/OpenOCRCorrect/FrameWorkCode$ ./qpadfinal 
QApplication: invalid style override passed, ignoring it.
Error:DictWords NOT Loaded

Error:GBookWords NOT Loaded in mapNvector

Error:IBookWords NOT Loaded in mapNvector
0 0
Error:PWordsWords NOT Loaded
0 words loaded in PWords
Error:CPair NOT there/loaded
0LSTM Pairs LoadedError Confusions NOT Loaded

Frontend log

Loading Dict Words
0Dict Words Loaded

 
Loading OCR Words
0 Indsenz Words Loaded & 
0 Google Doc Words Loaded.
 
 Loading Domain Words
0 Domain Words Loaded.
 
Loading Substrings and Sandhi Rules
0 Substrings and 71 Sandhi Rules Loaded.
 
Loading Confusions Please wait...

And then the application does not do anything.
What would I need to make it work?

How are getting the results in the graph of 'Avg. time per error' v/s 'Page No', like what is step by step procedure for obtaining the mentioned graph.

This link https://tinyurl.com/y7wbpo6m is broken which is mentioned on README

where can I get the dataset

I have read your research paper and I understood the overview of the project. Now, I have a doubt only in the following mentioned part from your research paper. Could you please tell us how to implement the below-mentioned part of the project and how to work with the data like f score etc. and training of the model. I have studied the theory about the logistic linear regression model but could not understand it's implementation in your project.

[ It can be observed that F-Score for the dual OCR system is better than LB (Lower Baseline) and are in-between (or better than) UB (Idealized Upper Baseline). For comparison purposes, we also trained a logistic linear regression based plug-in classifier [10] on 60% of data with 80:20 as the train:val split, and tested the results of remaining 40% of document words. The plug-in classifier is trained to optimize the F-score of validation data and can also be tuned incrementally on-the-fly. This is done by tuning the probability threshold of the logistic linear classifier to maximize the F-Score on validation data. Using the simple features like frequencies of n-grams (up-to 8) of the OCR word in a fixed dictionary, it gave better F-Score on the test data than Dual OCR agreement. ]

hello sir,
While using your test data of image and ocr text, after loading text for spell check the system perfomes well ,(for eg. misspelled words are represented with colour).

But when it comes for my test case, it automatically gets auto corrected without
suggesting wrong words with colour , and in that auto correction process some right words are also getting auto wrong.
Sir , could we operate system with CLA instead of GUI?
Thank you.