Retrieval from Large Document Image Collections

​

People involved: Riya Gupta, C.V. Jawahar ; Dataset & Server Handlers : Varun Bhargavan, Aradhana Vinod

Pipeline of our System

Dataset Snapshot

Demo Video

• An IRE system using Basic NLP techniques to extract information from ancient document collection in Indic languages.
• Documents are scanned, segmented using a line level segmentation algorithm and processed in the Indic-OCR and combined together to form this IRE system.
• Languages supported : Hindi, Tamil, Telugu, Sanskrit, Bangla
• Links to -
• Code
• Project website

​

Retrieval from Large Document Image Collections

​

**People involved: Riya Gupta*, Siddhant Bansal*, C.V. Jawahar

Dataset Statistics

Word Level Segmentations

Future Work :

• Using image preprocessing techniques & finetuning and self-training, improving the OCR accuracy.
• Generation of word Level Segmentations** on the large dataset without manual supervision and extracting the resulted visualizations using them.
• Enhancement of the IRE system semantically and contextually.

​

​

• Count of dataset : >1.5m doc. images
• Collaborators : NDLI (IIT KGP) & British Library
• Current Features :
• Query Level Search
• Language Level Search
• Relevance and Visualization
• Filtering
• Transliteration

Word Segmentation

Model