Applying OCR,Object Detection and NER techniques to Big Datasets

Objective: Apply Optical Character Recogntion, Named Entity Detection, Object Detection and Caption Generation on Big datasets

This project is partitioned into three parts:-

• Extracting data from scanned PDF files using OCR techniques
• Crawling and scraping ufostalker.com to get images and data; apply Object detection and captioning techniques to these images
• Apply NER techniques on data/sighting descriptions to extract different named entitites

Note: This project is built over the first one here

The dataset built from the first project is called v1 ufo dataset and the dataset generated from this project is called v2 ufo dataset.

Tools used

---

1. ImageMagick (convert scanned pdf to .tiff files)
2. Ghostscript (convert scanned pdf to .tiff files)
3. Poppler (to separate a pdf with multiple pages into single page pdfs)
4. Tesseract (for OCR from .tiff files)
5. Tika Dockers
6. TikaAndVission
7. Tika Image Captioning
8. Selenium (for scrapper/crawler)
9. OpenNLP
10. CoreNLP
11. NLTK
12. MITIE
13. Grobid

Measures of How to run some scripts

---

OCR

$> bash separate-pdf.sh /path/to/directory/with/pdf/files
$> bash pdftotext.ssh /path/to/directory/with/folders/for/pdf/files
$> bash extract-text.sh inputfile
$> bash extract_text.sh /path/to/directory/with/pdf/files output-directory language
$> bash ocr-pipeline.sh /path/to/directory/with/pdf/files
$> bash extract-ocr-final.py inputfile outputfile

Parsing

$> python3 clean_ocr.py inputfile outputfile 
$> python3 pythonParser2.py inputfile outputfile #text parser for british ufo and NER
$> python3 parser3.py inputfile outputfile #text parser for ufostalker and NER

Object-detection

$> python3 filter-relevant-images.py 
$> python3 get_images.py <file_containing_image_links> <output_dir>
$> python objects.py <input_dir>
$> python scraper.py <output_file>
$> python3 ufo_stalker_json.py
$> python3 version2.2-ufo.py objects-input-file data-inputfile outputfile
$> java ObjectRecognitionParser.java (with Tika)
$> python check_extensions.py <input_file> <output_file>

NER

$> python CoreNLP.py /path/to/dataset /path/to/ouputs
$> python OpenNLP.py /path/to/dataset /path/to/ouputs
$> python MITIE.py /path/to/dataset /path/to/ouputs
$> nltk-server -v --port 8881
$> python NLTK.py /path/to/dataset /path/to/ouputs 
$> mvn -Dmaven.test.skip=true jetty:run-war 
$> python Grobid.py /path/to/dataset /path/to/ouputs
$> python integrate_datasets.py /path/to/v1 /path/to/CoreNLP_results /path/to/OpenNLP_results /path/to/NLTK_results /path/to/MITIE_results /path/to/Grobid_results /path/to/outputs

Img2Text

$> python retrain.py --image-dir $dir_of_training_img

Merge datasets

$> python3 merge.py v1withNER-datasetfile britishUFO- datasetfile ufostalker-datasetfile outputfile(v2.json)
$> python3 jsonToTEV.py inputJSONfile outputTSVfile

Inferences

---

Refer to this document for inferences from this project.