[{"data":1,"prerenderedAt":44},["ShallowReactive",2],{"blog-post-one-newspaper-page-four-ocr-tools":3,"blog-index":14},{"slug":4,"title":5,"description":6,"date":7,"updated":7,"tags":8,"image":11,"html":12,"reading_minutes":13},"one-newspaper-page-four-ocr-tools","Tried OCR on an old newspaper? Compare this page","One dense 1850 newspaper page, four OCR tools: one nearly right, one fluent but wrong. See the results, then run your own page.","2026-10-20",[9,10],"ocr","text extraction","https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fog.088293acda.png","\u003Cp>If you have run an old scan through a free OCR site or a chatbot, you have probably seen one of two results. Either the text comes back full of broken words, or it comes back clean and readable, and only later do you notice it doesn't match the page.\u003C\u002Fp>\n\u003Cp>We built Presentr Analyze's text extraction to read dense, small-type pages the way they are printed. Rather than describe it, here is one hard page run through four tools, with every result checked against a human-proofread transcription. Then try a page of your own.\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fpage.09160a37d8.jpg\" alt=\"A scanned 1850 Scientific American page with four columns of small, densely set type.\" srcset=\"https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fpage.f663527af5.webp 640w, https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fpage.3ff9363679.webp 1024w\" sizes=\"(max-width: 768px) 100vw, 720px\" width=\"1120\" height=\"1600\" loading=\"lazy\" \u002F>\u003C\u002Fp>\n\u003Ch2 id=\"the-page-and-the-ground-truth\">The page and the ground truth\u003C\u002Fh2>\n\u003Cp>The source was page 171 of \u003Cem>Scientific American\u003C\u002Fem>, Series 1, Volume 5, Issue 22, published in 1850. It is a four-column page set in small type.\u003C\u002Fp>\n\u003Cp>We used the Wikimedia Commons scan as served by Wikisource at 1120 by 1600 pixels. The page is public domain. The scan has a “© 1850 Scientific American, Inc.” footer, which we did not treat as part of the transcription.\u003C\u002Fp>\n\u003Cp>Our ground truth was the \u003Ca href=\"https:\u002F\u002Fen.wikisource.org\u002Fwiki\u002FPage:Scientific_American_-_Series_1_-_Volume_005_-_Issue_22.pdf\u002F3\" rel=\"noopener\">Wikisource transcription of the page\u003C\u002Fa>. Wikisource marks it level 3, “proofread,” rather than “validated.” We excluded headers and kept printed misspellings as printed. The transcription is by Wikisource contributors and is available under \u003Ca href=\"https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby-sa\u002F4.0\u002F\" rel=\"noopener\">CC BY-SA 4.0\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch2 id=\"what-we-tested\">What we tested\u003C\u002Fh2>\n\u003Cp>On October 20, 2026, we gave the image to each of these tools:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Presentr Analyze’s extraction, which uses Anthropic’s Claude Sonnet 4.6 model.\u003C\u002Fli>\n\u003Cli>OpenAI gpt-4o through the API, with image detail set to “high.”\u003C\u002Fli>\n\u003Cli>OpenAI gpt-5 through the API.\u003C\u002Fli>\n\u003Cli>Tesseract 5 with the \u003Ccode>tessdata_best\u003C\u002Fcode> English model. Tesseract is the open-source engine behind many free online OCR sites.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>These were API runs, not runs in the ChatGPT app. We ran each tool once. Where a tool accepted an instruction, we used the same one: “Extract all text visible in this image verbatim. Return only the text with no preamble or commentary.”\u003C\u002Fp>\n\u003Cp>We measured character error rate: the edit distance between an output and the ground truth, divided by the ground-truth length. Before calculating it, we collapsed whitespace and rejoined words that had been hyphenated across lines.\u003C\u002Fp>\n\u003Ch2 id=\"results\">Results\u003C\u002Fh2>\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Tool\u003C\u002Fth>\n\u003Cth style=\"text-align:right\">Character error rate\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>Presentr Analyze\u003C\u002Ftd>\n\u003Ctd style=\"text-align:right\">2.9%\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Tesseract 5\u003C\u002Ftd>\n\u003Ctd style=\"text-align:right\">18.1%\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>gpt-4o\u003C\u002Ftd>\n\u003Ctd style=\"text-align:right\">61.8%: fluent sentences that are not on the page\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>gpt-5\u003C\u002Ftd>\n\u003Ctd style=\"text-align:right\">Declined: “I can’t extract the text because it’s too small\u002Flow‑resolution to read accurately. Please upload a higher-resolution image or a zoomed\u002Fcropped section.”\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003Cp>On this page, Analyze produced a 2.9% character error rate, compared with 18.1% for the tested Tesseract workflow. The tested gpt-4o workflow returned fluent but mostly wrong text. The tested gpt-5 workflow declined to extract the page.\u003C\u002Fp>\n\u003Ch2 id=\"one-passage-four-outputs\">One passage, four outputs\u003C\u002Fh2>\n\u003Cp>The difference is clearer in one aligned passage.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Ground truth\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>No foreign magazine contains such minute information upon those subjects, and to our knowledge we have saved thousands of dollars to our country, and saved a great waste of ingenuity. We are now illustrating the history of propellers.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Cstrong>Presentr Analyze\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>No foreign magazine contains such minute information upon those subjects, and to our knowledge we have saved thousands of dollars to our country, and saved a great waste of ingenuity. We are now illustrating the history of propellers.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Cstrong>Tesseract 5\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>No fomign maga- ine contains suck minute information upea| thee subject, and to cur knowledge we have: coved thowands of dollaes tocur country, ant feved a real waste of ingenaity. We ave now illostrating the history of propellers.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Cstrong>gpt-4o\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>The most exquisite minie informations may be obtained in the somlicy known, for the lowed thousand of dollars in our country, and express! the claims of ingenuity.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Cstrong>gpt-5\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>I can’t extract the text because it’s too small\u002Flow‑resolution to read accurately. Please upload a higher-resolution image or a zoomed\u002Fcropped section.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fpassage.801390a233.jpg\" alt=\"Enlarged crop of the same passage from the scan: "No foreign magazine contains such minute information upon those subjects, and to our knowledge we have saved thousands of dollars to our country, and saved a great waste of ingenuity."\" srcset=\"https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fpassage.685181993b.webp 640w\" sizes=\"(max-width: 768px) 100vw, 720px\" width=\"726\" height=\"300\" loading=\"lazy\" \u002F>\u003C\u002Fp>\n\u003Cp>Tesseract’s mistakes are visible: broken words, misspellings, and stray punctuation tell us that the text needs checking. The gpt-4o excerpt reads more smoothly, but it is not the passage on the page. That is the more dangerous failure because it reads as if it were right.\u003C\u002Fp>\n\u003Ch2 id=\"what-this-page-does-and-does-not-show\">What this page does and does not show\u003C\u002Fh2>\n\u003Cp>This is one page and one run per tool. The comparison used APIs rather than consumer apps, and its ground truth was proofread but not validated. Different settings, prompts, or newer models may produce different results.\u003C\u002Fp>\n\u003Cp>In a small exploratory test of 10 pages covering handwriting, printed Hindi, and multi-column print, the widest gap between Analyze and the other tools was on dense multi-column print. On a cleaner two-column page, Tesseract did slightly better than Analyze (1.6% against 2.0%). That is broader context, not a benchmark.\u003C\u002Fp>\n\u003Cp>Handwriting, especially old scripts, remains difficult. Every tool in our test, including ours, still makes serious mistakes on it. Our extraction page tells users to check names and dates.\u003C\u002Fp>\n\u003Ch2 id=\"run-your-own-page\">Run your own page\u003C\u002Fh2>\n\u003Cp>The best comparison is your own material. \u003Ca href=\"https:\u002F\u002Fanalyze.presentr.me\u002Ftools\u002Fextract-text\" rel=\"noopener\">Upload a page, PDF or photo to Analyze\u003C\u002Fa> and put the result next to whatever you use now. You can read it page by page, copy it, or download it as text.\u003C\u002Fp>\n\u003Cp>A new account starts with 5 credits: enough for two PDF pages or five images. After that, extraction costs 2 credits per PDF page and 1 per image.\u003C\u002Fp>\n\u003Cp>The full outputs from this test are available on request.\u003C\u002Fp>\n",5,{"site":15,"generated_at":16,"posts":17,"redirects":43},"analyze","2026-10-20T20:55:10.161Z",[18,22,33],{"slug":4,"title":5,"description":6,"date":7,"updated":7,"tags":19,"image":11,"url":20,"published_at":21},[9,10],"https:\u002F\u002Fanalyze.presentr.me\u002Fblog\u002Fone-newspaper-page-four-ocr-tools","2026-10-20T20:50:29.423Z",{"slug":23,"title":24,"description":25,"date":26,"updated":26,"tags":27,"image":30,"url":31,"published_at":32},"fix-progressive-subtitle-drift","Subtitles Offset or Slowly Drifting? Diagnose the Difference","Check subtitle sync at the beginning, middle, and end to distinguish a constant offset, progressive drift, or a re-edited video.","2026-10-19",[28,29],"subtitles","captions","https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fog.743e8292aa.png","https:\u002F\u002Fanalyze.presentr.me\u002Fblog\u002Ffix-progressive-subtitle-drift","2026-10-19T19:30:48.590Z",{"slug":34,"title":35,"description":36,"date":37,"updated":37,"tags":38,"image":40,"url":41,"published_at":42},"subtitle-file-out-of-sync","A subtitle file can look valid and still be out of sync","A subtitle file 2.5 seconds late scored the same as the in-sync one on our own check. Why a text check cannot see sync, and what the audio shows.","2026-10-15",[28,29,39],"video","https:\u002F\u002Fcontent.presentr.me\u002Fanalyze\u002Fassets\u002Fog.f3a05ee827.png","https:\u002F\u002Fanalyze.presentr.me\u002Fblog\u002Fsubtitle-file-out-of-sync","2026-10-15T15:51:57.159Z",{},1792698413571]