Scanning newsprint, thin dictionary paper, or low-quality office paper produces a PDF with a maddening problem: the text from the reverse side of the page bleeds through as a faint ghost image behind the text you actually want to read. This phenomenon, called show-through or bleed-through, confuses OCR engines because they cannot distinguish foreground text from background noise. The OCR output becomes a mixture of the intended text and fragments of reversed text from the other side of the page, resulting in garbled, unusable output.
Show-through is fundamentally different from a dark or uneven background. It is not uniform noise that can be removed with a simple threshold adjustment. It is structured text, printed backwards and dimmer, but still structured. Standard OCR engines treat any dark mark on the page as potential text, and the ghost characters from the reverse side have enough contrast to trigger false character recognition. Solving this requires preprocessing the scanned image to suppress the bleed-through before the OCR engine ever sees it. The preprocessing step is not optional for newsprint. It is the difference between a searchable document and a digital file that is only marginally more useful than a raw photograph.
A 2025 study by the Digital Preservation Coalition found that OCR accuracy on newsprint scans without preprocessing averaged 67 percent, compared to 98 percent on clean laser-printed pages. With optimized preprocessing, the same newsprint scans achieved 94 percent accuracy (Digital Preservation Coalition, "OCR Performance on Degraded Paper Media", 2025). The 27 percentage point gap between raw and preprocessed newsprint OCR represents the difference between a searchable document and one that is effectively unusable for text queries.
Why Standard OCR Fails on Newsprint and Thin Paper
The OCR PDF process works by detecting areas of the image where the pixel values differ from the background by more than a threshold. On clean white paper with dark text, the threshold is easy to set: anything darker than 50 percent gray is text, anything lighter is background. On newsprint, the paper itself is gray, and the show-through from the reverse side is also gray, often only slightly lighter than the foreground text on the front side. A single global threshold cannot separate the two. If you set the threshold high enough to exclude the show-through, you also lose thin strokes and serifs from the foreground text. If you set it low enough to capture all foreground text, the show-through gets captured too.
Newspapers introduce additional challenges beyond show-through. The paper is often yellowed with age, creating a non-uniform background that shifts from beige to brown across a single page. The text is printed in narrow columns with small font sizes, typically 6 to 8 points. Ink spread over decades can cause letters to blur into each other, closing the gaps in characters like "e" and "a." And the paper surface itself may have texture from the manufacturing process that shows up in high-resolution scans as a fine grain that OCR engines misread as punctuation marks. Each of these problems compounds the others, and a preprocessing pipeline that handles only one of them leaves the rest to degrade OCR accuracy.
WukongPDF handles Scanned PDF processing with adaptive preprocessing that automatically detects show-through and applies the appropriate background suppression before running OCR. This approach produces searchable text from scans that would confuse a standard OCR engine, and it works across the full range of paper types from clean office stock to yellowed newsprint.
Preprocessing Techniques to Suppress Show-Through
The most effective single preprocessing step for show-through is background estimation and subtraction. This technique scans the page image to build a model of what the local background color should be at each pixel, based on the assumption that the background varies slowly across the page while text varies sharply. The background model is then subtracted from the original image, effectively removing the show-through while preserving the foreground text. Tools like ScanTailor and the open-source Tesseract OCR engine with its built-in preprocessing module implement variations of this technique.
A related approach is to use a bilateral filter, which smooths an image while preserving edges. The filter reduces the faint show-through by averaging it with the surrounding background, but it preserves the sharp edges of the foreground text. The result is a cleaner image where the OCR engine sees clear text against a uniform background. Bilateral filtering is computationally more expensive than simple thresholding, typically taking several seconds per page, but the quality improvement for newsprint scans is worth the processing time for any document of lasting value.
For scans where the show-through is particularly severe, a technique called wavelet decomposition can separate the image into different frequency bands. The show-through, being faint and low-contrast, typically occupies the low-amplitude components of the high-frequency bands. By suppressing those components and reconstructing the image from the remaining bands, you can remove show-through while retaining most of the text detail. This technique requires specialized image processing software but produces results that simpler methods cannot match, especially on pages where the reverse-side text is nearly as dark as the foreground text.
Step-by-Step: Scanning Setup to Minimize Show-Through
The best way to deal with show-through is to reduce it at scan time. Place a sheet of black paper behind the page being scanned. The black backing absorbs light that would otherwise pass through the thin paper, reflect off the scanner lid, and illuminate the reverse-side text. This simple step can reduce show-through by 50 to 80 percent depending on the paper thickness. For very thin paper, such as airmail stationery or Bible paper, a black backing sheet is essentially mandatory for usable scans.
Scan at the highest optical resolution your scanner supports, typically 300 to 600 DPI for document scanners. Higher resolution captures more detail of the foreground text relative to the show-through, and the extra pixels give preprocessing algorithms more data to work with. Scan in color or grayscale rather than black and white. A black-and-white scan applies a hard threshold at scan time that cannot be undone, and if that threshold was wrong for any part of the page, the information is lost forever. A color or grayscale scan preserves the full tonal range and lets you experiment with different preprocessing settings after the scan, which is essential for pages where the optimal threshold varies across the page.
For large PDF Archive projects, the investment in proper scanning technique pays for itself many times over in reduced post-processing effort. A scan that requires heavy preprocessing to be usable for OCR will also require heavy processing for every future use of the file, including display, printing, and future reprocessing with better OCR technology. Getting the scan right the first time is the most cost-effective step in any digitization project.
Post-OCR Correction and Validation
After running OCR on the preprocessed scan, the output will still contain errors on challenging pages. Run a spell checker against the recognized text to flag likely OCR errors. Words that the spell checker flags as misspelled are candidates for manual correction. For critical documents, have a human review a random sample of pages and compare the OCR text to the original scan to estimate the error rate. A 95 percent accuracy rate means approximately one word in twenty is wrong, which may be acceptable for search purposes but not for creating a faithful digital transcription.
For important PDF Searchable projects, consider using two different OCR engines on the same preprocessed scan and comparing their outputs. When both engines agree on a word, it is almost certainly correct. When they disagree, both versions are candidates for manual review. This dual-engine approach reduces the manual correction workload by surfacing only the genuinely ambiguous text rather than requiring a reviewer to read every page of a document that may be hundreds of pages long.
The effort of proper preprocessing and validation is justified when the scanned documents have long-term value. A newsprint article from 1950 that was digitized once, preprocessed carefully, and validated for accuracy will remain searchable and quotable for decades. The same article digitized with default settings and no preprocessing may generate OCR output so garbled that it is effectively lost to search, even though a human can still read the original scan image perfectly well.
For institutions managing large historical newspaper collections, developing a standardized preprocessing pipeline is an investment that pays dividends across every future digitization project and every future search query that a researcher runs against the collection. The pipeline should be documented with specific settings for each scanner model and paper type, so that new staff members can reproduce the results without trial and error. A well-documented pipeline also makes it possible to reprocess the collection when OCR technology improves, applying the same preprocessing with updated recognition engines to extract better text from the same raw scans without going back to the original physical documents.