Full Page OCR
Can anyone Help me to perform OCR on full page Handwritten Text, This model works great if the image only contains one line of text
Some people suggested to use text detectors like CRAFT to get the text regions and then send them to trOCR. I'm still managing to get this working though
I have tried using using CRAFT too but it is failing
Thanks for the reply 😊
I try it to use cv2 to get where texts parts are, then cropp it as a one line text
I am using easyOCR detect method to get the list of bounding boxes, grouping them in line with some logic and creating cropped ROI's of those boxes to pass into the model one by one hope it helps.
I used the 35th to 100th characters encoded in Unicode and converted them into images, and used this model for recognition. However, few of the models were able to recognize them correctly, and even the letter g was recognized as two 9s