Croping document pages images

Deep Learning
1

Hi everyone ,
I am trying to build model that crops the scanned pages of documents correctly , i have tried to anotate 2K images with bounding box and trained using fastrcnn and YOLOv8 but those are not much accurate on testing . is there any other model which can be finetuned for this task to do better accuracy or any other preprocessing should i have done ??any link or any suggestions would be helpful

2

Hi @chandrakanta,

I haven’t used those models, but I did some object detection tasks with YOLOv10, and the results were promising. I mainly used their GitHub repository and read the paper to get a sense of the model’s functionality and how to use it. In addition to that, I used Roboflow Universe for the dataset. You might find larger and higher-quality datasets there!

And if you haven’t considered yet, these approaches could be helpful too:

  • Augmenting images using techniques like rotation, scaling, and contrast adjustment
  • Normalizing and standardizing your images
  • Implementing edge detection algorithms to enhance document edges
3

Thanks a lot , I have not tried YOLOv10 yet , but have tried with contrast and image augumentation,let me do try with edge detection and yolov10 and will let you know if it helps …

1 Like