OCR0048: Enhance Synthetic Page Images with Augmentation Libraries

[jim-gyas]

Description:

Develop a Python script to apply augmentations to synthetic page images using various libraries. Users can select their preferred augmentations from a predefined list. The script will enhance the existing synthetic page images by applying the selected augmentations. The main goal is to create diverse and realistic training images to improve the OCR dataset.

Resources:

1)Synthetic Page Image: https://github.com/OpenPecha/SynthImage 2)Augmentation Libraries:

• Albumentations

• Augraphy

  List of Augmentation using Albumentations:

  • Blur
    • Blur
    • Median Blur
    • Motion Blur
  • Flip
    • Vertical Flip
    • Horizontal Flip
  • Grid Distort
  • Hue_Saturation
  • Perspective
  • Random Rain
  • Random Shadow
  • Rusting
  • Solarize
  • Sun flare
  • Transpose

List of Augmentation using Augraphy:

• BadPhotoCopy
• InkBleed
• PaperTexture
• DirtyRollers
• LowInkPeriodicLines
• Scribbles
• WaterMark
• Faxify

Implementation:

Sub Task

• [x] For each augmentation( Albumentation Library), create a module within SynthImage Folder/Directory.
• [x] Test each Albument augmentation to ensure correct augmentation and output saving.
• [x] For each augmentation(Augraphy Library), create a module within SynthImage Folder/Directory.
• [x] Test each Augraphy augmentation module to ensure correct augmentation and output saving.

[jim-gyas]

Sythetic Page Image

Albumentation Augmentation:

Perspective Augmentation

Median Blur Augmentation

Shadow Augmentation

Sun Flare

[jim-gyas]

Original Image:

Augraphy Albumentation Example:

[kaldan007]

@jim-gyas I would rather make each augmentation a separate module like you have done with distrotion and deformation. selection of with augmentation is common to all augmentation. Then why do u have that selection feature only for those two libraries augmentation. I remember u have a randomization feature implemented in a module.

[jim-gyas]

Transpose Augmentation:

Problem in Line Extraction for Transpose Augmentation:

After Solving the problem:

[jim-gyas]

Input Synthetic Page Image:

Augraphy Augmentation:

Bad Photo Copy Augmentation:

Dirty Rollers Augmentation:

Ink Bleed Augmentation:

Scribble Augmentation:

Page Texture Augmentation: