hocr2pdf

Takes a hocr file (output from the likes of Tesseract / Omnipage / ABBYY FineReader) and merges with an image to create a searchable PDF file. As of 30/08/2015 it will also take advantage of the textangle value in ocr_line class span's when processing to convert words/bbox's to their correct orientation.

I think only newer builds of Tesseract will utilise this correctly, run something like the following to generate a sample hocr file:

> tesseract.exe infile.tif outfile -psm 1 hocr

Use pdfkit to draw PDF files.

Utilises cheerio for HTML parsing.

var fs = require('fs')
var PDFDocument = require('pdfkit')
var sharp = require('sharp') // http://sharp.dimens.io
var hocrConv = require('./index.js')

// Create new hocrConv object with hocr file
var hocrObj = new hocrConv(fs.readFileSync('hocr_output.html'))
var image = fs.readFileSync('image.png')

var options = {
  size: [
    hocrObj.width,
    hocrObj.height,
  ],
  margin: 0  // you want a margin of zero
}

//start a new PDF doc
var pdfDoc = new PDFDocument(options)
var stream = fs.createWriteStream('output.pdf')
pdfDoc.pipe(stream)

// Open up the image file in sharp
sharp(image)
  .rotate(hocrObj.rotation)
  .toBuffer(function(err, data, info) {
    if (err) { throw err }

    // this allows doing manipulations of the image before inserting it into the PDF document

    // generate the words
    var words = hocrObj.getWords()
    for (var x = 0; x < words.length; x++) {
      // use height of bbox
      pdfDoc.fontSize(words[x][3])
      pdfDoc.font('Helvetica')
      pdfDoc.fillColor('black')
      pdfDoc.text(words[x][4], words[x][0], words[x][1]), {
        width: words[x][2],
        height: words[x][3],
        aligh: 'centre'
      }
    }

    // optional: draw bounding boxes around paragraphs and/or words
    var wordboxes = hocrObj.drawWordBoxes()
    for (var w = 0; w < wordboxes.length; w++) {
      pdfDoc.rect(wordboxes[w][0],wordboxes[w][1],
                  wordboxes[w][2],wordboxes[w][3])
            .strokeColor('#ADD8E6')
            .fillColor('#ADD8E6')
            .lineWidth(0)
            .fillOpacity(1)
            .stroke()
    }

    var paraboxes = hocrObj.drawParagraphBoxes()
    for (var p = 0; p < paraboxes.length; p++) {
      pdfDoc.rect(paraboxes[p][0],paraboxes[p][1],
                  paraboxes[p][2],paraboxes[p][3])
            .strokeColor('#90EE90')
            .lineWidth(0.5)
            .fillColor('#90EE90')
            .fillOpacity(0.5)
            .dash(6, {space: 3})
            .stroke()
    }

    pdfDoc.image(data, 0, 0, {
      fit: [hocrObj.width, hocrObj.height]
    })

    // finalise the PDF
    pdfDoc.end()
  }
)
//thats it.

Note that the PDF canvas creates a new layer every time you draw / write / insert an image. So the order you do it in matters.

Oh, and:

var dpi_of_original_image = 300

hocrObj.findPageSize(dpi_of_original_image)
// will return an array with width, height, name & orientation of page size
// ie [595, 842, 'A4', 'Portrait']

console.log(hocrObj.rotation) // detected rotation
console.log(hocrObj.dmns) // raw dimensions array from searching hocr file
console.log(hocrObj.width)
console.log(hocrObj.height)
console.log(hocrObj.hocr) // cheerio object for doing DOM traversal etc.

Yeah I'm not sure what I needed that for (findPageSize). It might come in handy later?