Lua Requires Chilkat v11.7.0+

Assess PDF Text Quality Before Extracting

See more PDF Text Extraction Examples

Demonstrates the "assess" operation, which returns only the quality report: whether the text can be trusted, which pages are scanned images that need OCR, and how each page's text was found. Use it to triage documents before extracting. (Chilkat does not perform OCR.) Lua

local chilkat = require("chilkat")

-- This example requires the Chilkat API to have been previously unlocked.
-- See Global Unlock Sample for sample code.

-- The PDF used in this example (page 2 is a scanned image with no text) is available at mixed_text_image.pdf

local pdf = chilkat.Pdf.new()
local success = pdf:load_file("qa_data/pdf/mixed_text_image.pdf")
if not success then
    print(pdf.last_error_text)
    return
end

local json = chilkat.JsonObject.new()
local jsonOut = chilkat.JsonObject.new()

-- "assess" computes the quality report without producing any text.
-- It is fast, and it is always allowed, even for an encrypted PDF that forbids copying.
success = pdf:extract_text("assess", json, jsonOut)
if not success then
    print(pdf.last_error_text)
    return
end

-- The overall rating is the worst rating among the pages that have text:
-- "good", "fair", "poor", or "none" when there is no text at all.
print("Rating: " .. jsonOut:string_of("quality.rating"))

-- Is the PDF tagged (does it have a structure tree)?
print("Tagged: " .. tostring(jsonOut:bool_of("quality.tagged")))

-- The pages that are images only and would need OCR, such as "2" or "4,9".
print("Pages needing OCR: " .. jsonOut:string_of("quality.pagesNeedingOcr"))

-- The per-page report.
local j = 0
local count_j = 0
local i = 0
local count_i = jsonOut:size_of_array("quality.pages")
while i < count_i do
    jsonOut.i = i
    -- textSource is "text" (real text), "ocrLayer" (invisible text over an image),
    -- "imageOnly" (needs OCR), "mixed", or "empty".
    print("Page " .. jsonOut:int_of("quality.pages[i].page") .. ": rating " .. jsonOut:string_of("quality.pages[i].rating") .. ", textSource " .. jsonOut:string_of("quality.pages[i].textSource"))
    -- Each issue has a machine-readable code and a sentence.
    j = 0
    count_j = jsonOut:size_of_array("quality.pages[i].issues")
    while j < count_j do
        jsonOut.j = j
        print("    " .. jsonOut:string_of("quality.pages[i].issues[j].code") .. ": " .. jsonOut:string_of("quality.pages[i].issues[j].message"))
        j = j + 1
    end

    i = i + 1
end

-- Decide what to do with the document:
if jsonOut:string_of("quality.rating") == "none" then
    print("No extractable text; the whole document needs OCR.")
elseif jsonOut:string_of("quality.pagesNeedingOcr") ~= "" then
    print("Some pages need OCR; the remaining pages can be extracted with the text or markdown operation.")
else
    print("All pages can be extracted.")
end

-- Sample output:

-- Rating: good
-- Tagged: false
-- Pages needing OCR: 2
-- Page 1: rating good, textSource text
-- Page 2: rating none, textSource imageOnly
--     imageOnly: No text; page is a scanned image
-- Page 3: rating good, textSource text
-- Some pages need OCR; the remaining pages can be extracted with the text or markdown operation.