Assamese Newspaper OCR — Digitizing Print Archives

Why Newspaper OCR Is Not Just Document OCR

A clean book scan and a newspaper page are different OCR problems. Books have consistent fonts, predictable column widths, and relatively clean paper. Newspapers are the adversarial case: multiple font sizes on a single page, narrow columns separated by hairline rules, newsprint that textures scans with speckle noise, and ink that fades at different rates across the page depending on original press coverage.

For Assamese newspapers specifically, these generic newspaper challenges combine with Assamese script complexity — the Shirorekha top bar, the Juktakkhor conjunct ligatures, the matra positioning above and below — creating a recognition task that requires purpose-built software. DRISTI OCR is built specifically for this environment.

Multi-Column Layout Analysis

A standard Assamese daily newspaper page uses 5–8 columns of varying widths. Headline stories span multiple columns; single-column body text runs beside them. Classified ad sections use a dense 8-column grid. Sports results tables appear in a separate layout zone.

OCR engines that treat a newspaper page as a single text block produce output where column 1 content is interleaved with column 2 content — because the scanner reads left to right across the full page width rather than down each column. DRISTI's layout analysis step detects column boundaries, identifies spanning headlines, and establishes a reading order before recognition begins. Output text follows the correct editorial reading order.

Column Boundary Detection

Assamese newspaper columns are separated by whitespace or thin vertical rules. DRISTI detects both. For pages where the rule has faded or where columns nearly touch (a common condition in older tabloid-format papers), the engine uses text density analysis to infer boundaries rather than relying solely on visible whitespace.

Mixed Fonts Across a Single Page

Before approximately 2005, Assamese newspapers typically used one font family for body text and a different family — often a visually heavier display variant — for headlines. Both were typically Geetanjali-family fonts but from different font files with different glyph designs. After 2010, some papers introduced Unicode fonts for specific sections while retaining Geetanjali for others.

This means a single page may require multiple recognition models applied to different zones. A recognition engine trained on a single font model will produce systematic errors in every zone where the font does not match. DRISTI applies zone-specific model selection, improving accuracy across mixed-font pages compared to single-model engines.

Newsprint Texture and Ink Fading

Newsprint paper is porous. Ink bleeds slightly into the paper fiber, producing rounded stroke edges and occasional inter-character ink bridges. The binarization step — converting a grayscale scan to black and white — must handle this correctly. Too aggressive a threshold removes faint strokes; too permissive a threshold creates ink bridges between adjacent characters.

DRISTI applies adaptive binarization that locally adjusts the threshold based on neighborhood pixel statistics. This allows it to recover faint ink in one part of the page while avoiding ink bridge creation in a denser, darker section. For archive material from before 1990 — where newsprint has yellowed and ink has faded unevenly — scan at 400 DPI and let DRISTI's automatic preprocessing handle the binarization. See the complete Assamese image to text guide for preprocessing detail.

Classified Ad Sections

Classified advertisements are the highest-density, smallest-font sections of any Assamese newspaper page. Font sizes of 6–8pt at print resolution correspond to approximately 25–35 pixel character heights at 300 DPI. At this scale, the distinction between similar Assamese characters — ন vs ণ, শ vs স — approaches the resolution limit.

For classified sections, scanning at 400 DPI instead of 300 DPI measurably improves accuracy by increasing character height to 35–45 pixels — above the threshold where fine stroke distinctions are reliably preserved. Alternatively, configure DRISTI to apply a different recognition threshold for low-confidence zones and flag classified section output for additional proofreading.

Batch Processing Newspaper Archives

For archive digitization projects — converting decades of bound newspaper volumes to searchable digital format — DRISTI's batch mode is the practical path. DRISTI processes approximately 250 pages per 10 minutes, meaning a year of a daily newspaper (approximately 4000 pages including supplements) can be processed in under 3 hours of OCR time.

The project timeline is dominated by scanning and proofreading, not recognition. A flatbed scanner with automatic document feeder running at 400 DPI produces roughly 20–40 pages per hour; a year of newspaper pages requires 100–200 scanning hours. Plan the scanning operation before OCR capacity — scanner throughput, not recognition speed, is the binding constraint for large archive projects.

Recommended Workflow for Newspaper Archive OCR

1. Scan at 400 DPI in grayscale TIFF format
2. Apply automatic deskew before batch loading into DRISTI
3. Configure per-zone settings: minimum font size for classified sections, zone detection sensitivity for narrow-column layouts
4. Run DRISTI batch OCR on the full folder of page images
5. Export as searchable PDF (overlays recognized text on the original page image) and plain Unicode text
6. Proofread body text zones at a sampling rate appropriate to your accuracy requirements (10–20% page sampling for archive, 100% for reprinting)

For a full comparison of OCR tools available for Assamese newspaper digitization, see the best Assamese OCR software comparison.

After digitizing newspaper text, the output is Unicode. If your newspaper production pipeline uses PageMaker with Geetanjali fonts, convert the Unicode OCR output with Rupantarak before importing into your layout. For the complete operational picture of how a working Assamese newsroom handles the Unicode–Geetanjali barrier every production cycle, read the Assamese newspaper DTP workflow article. For typing new content rather than digitizing it, see the Jahnabi Pro Keyboard.