Image preprocessing and templates PPT.pdf

Ppt for image processing

1. INTERNSHIP REPORT -PPT PRESENTATION
IMAGE CLEANING & MODEL EXPLORATION
Kaviya . R
B Tech AI-DS (IV year)

2. Introduction
Focus: Cleaning engineering diagrams by removing unwanted text, logos,
and tables.
Explored manual and AI-based methods.
Worked with OCR (Text Detection) and Inpainting (Text Removal)
techniques.
Datasets: ~2800 images cleaned and processed.

3. Tools & Techniques
OCR Tools: EasyOCR, Tesseract, CRAFT, TrOCR, DBNet.
Image Cleaning: OpenCV Inpainting, ZITS (deep inpainting).
Models Used: ClipViT, ResNetViT, ConvNeXt Autoencoder.
Manual Editing: MS Paint for dataset preparation.

4. Key Work Done
Implemented multiple OCR methods for text recognition.
Automated text removal using EasyOCR + OpenCV inpainting.
Explored deep learning pipelines (DBNet + ZITS).
Trained and tested hybrid models (ResNetViT, ClipViT, ConvNeXt).
Tuned loss functions (BCE) to improve accuracy.
Compared outputs of different models.

5. Results

6. Conclusion
Explored multiple approaches for image cleaning.
Final models identified and tested:
ResNet-ViT
CBAM
UNet
ClipViT
These models gave the most promising results for preserving diagrams
while removing text and unwanted elements.