charmichokshi / optical-character-recognition

Detecting boxes and extract them one by one accurately for all PDFs of invoice to extract text within it like consignee and shipper name & address.

Optical-Character-Recognition

Detecting boxes and extract them one by one accurately for all PDFs of invoice to extract text within it like consignee and shipper name & address.

Algorithm:

• Convert PDF to PNG
• Apply morphological operations. I have defined two kernels.
  1. Kernel to detect horizontal lines.
  2. Kernel to detect vertical lines.
• Add the output of this two images which will detect all the boxes
• Crop all boxes which have height and width in a specific range and apply OCR on them to detect text

Naming convention:

• input: should have the input PDFs
• outputPNG: it will store PNG with respect to every PDF file
• Cropped: contains cropped images after applying morphological operations and text file generated after applying OCR

Input:

Cropped boxes

Extracted Consignee's Text

M/S JEENA & COMPANY JEENA HOUSE PLOT 170 OM NAGAR OFF PIPELINE ROAD SAGAR ANDHERI E MUMBAI 400 099 INDIA MR. SHIRISH BAADKAR 91 22 44222111

Extracted Shipper's Text

TRANSGROUP INTERNATIONAL 140 Eastern Ave Chelsea, MA 02150 [phone]