History of Scanning

History

Scanning has three deep roots that predate computers: facsimile/telegraphy (transmitting images over wire), photomechanical prepress (color-separating images for printing), and digital computing. Modern scanning fused these lineages.

Facsimile and photoelectric precursors (19th-early 20th c.). The conceptual ancestor of scanning is the fax machine. Alexander Bain patented an early facsimile process in 1843. Frederick Bakewell later improved the design, replacing Bain's pendulums with rotating cylinders (circa 1848) and publicly showing a facsimile apparatus at the 1851 Great Exhibition in London — mechanically an early drum-scanning arrangement, though it sensed conductivity rather than light. Around 1902, Arthur Korn's phototelautograph introduced a light-sensitive selenium cell, moving the concept toward true photoelectric image capture.

Analog color drum scanners for print (1937 onward). Alexander Murray and Richard Morse at Eastman Kodak built an early analog color drum scanner in 1937, using photocells behind red, green, and blue filters to read color originals for the printing industry. Printing Developments Incorporated (P.D.I.) later adapted the design and applied photomultiplier tubes (PMTs) to produce amplified signals for color-corrected printing values. These machines were prepress tools, not computer peripherals.

The first digital image capture (1957). The pivotal moment for digital scanning came in 1957 at the U.S. National Bureau of Standards (NBS, now NIST). A team led by Russell A. Kirsch built a rotating-drum scanner coupled to SEAC (the Standards Eastern Automatic Computer) and scanned a head-and-shoulders photograph of Kirsch's infant son, Walden. NIST records the image as measuring 176 pixels on a side, roughly 5 cm by 5 cm, grainy and black-and-white, and describes it as the first image scanned into a computer and stored as digital data. The distinction matters: analog drum scanners existed decades earlier for prepress, so 1957 marks digital storage of a scanned image, not the first drum scanner.

The Kurzweil reading-machine era (mid-1970s). In the 1970s, Raymond Kurzweil and Kurzweil Computer Products combined omni-font optical character recognition (OCR), a flatbed scanner, and text-to-speech synthesis into the Kurzweil Reading Machine for the blind, unveiled on 13 January 1976. Two commonly repeated "first" claims should be hedged: omni-font OCR was already in use earlier (for example, by CompuScan in the late 1960s-1970s), so Kurzweil is not the sole originator; and the attribution of the first CCD flatbed scanner to the Kurzweil machine (around 1976-1977) is asserted by secondary sources but is not firmly established in primary documentation — treat it as disputed.

Desktop flatbed scanners and desktop publishing (1980s). Early flatbeds such as the Autokon 8400 (1975) targeted print image processing. Personal-computer devices followed, including the ThunderScan scanning cartridge for the Macintosh (1984) and the Datacopy Model 700 (early 1985), described as the first flatbed scanner for the IBM PC. The breakout consumer product was Hewlett-Packard's HP ScanJet, released in 1987: a 300 dpi, 4-bit (16 gray levels) monochrome flatbed connectable to PCs and Macs, designed to complement HP's LaserJet printers in the desktop-publishing market. Per a Gartner Dataquest figure cited by secondary sources, it reportedly reached about 27% of scanner sales by dollar volume by early 1988; the 1989 ScanJet Plus kept the 300 dpi resolution but raised bit depth to 8 bits (256 gray levels).

Standardization and the driver problem (1990s). As scanners proliferated, incompatible proprietary drivers became a barrier. The TWAIN Working Group formed in 1992 (founding members Aldus, Caere, Hewlett-Packard, Eastman Kodak, and Logitech) to create a free, open standard connecting applications and imaging devices; TWAIN 1.0 shipped the same year. On Unix and Linux, the SANE (Scanner Access Now Easy) project emerged in the late 1990s to provide an open, hardware-abstracting API.

Multifunction devices and networked scanning (2000s-present). Scanning migrated from standalone peripherals into multifunction printers (print/scan/copy/fax in one chassis) and onto the network. Standards bodies including the Printer Working Group (PWG) and the Mopria Alliance defined driverless network scanning protocols — notably eSCL (the scan counterpart to Apple's AirPrint, published by Mopria) and PWG IPP-based scanning — allowing a client to scan without a vendor-specific driver.

How it works

At its core, a scanner performs the same operation as the 1957 NBS machine: sample light intensity at points across the original and digitize each sample.

• Illumination. A light source (historically a cold-cathode fluorescent lamp; now typically LEDs) illuminates the original. Reflective originals such as paper and photos are lit from the same side as the sensor; transparent originals such as film and slides require a transparency unit that lights from behind.
• Optical path. For CCD scanners, a moving carriage of mirrors and a lens focuses reflected light from a scan line onto a linear image sensor. For contact image sensor (CIS) scanners, a row of sensors sits directly beneath the glass with no long optical path.
• Sensing. The sensor converts light to an analog electrical signal proportional to intensity. Color is obtained either by filtering the sensor into red, green, and blue channels or by sequential passes or illumination.
• Analog-to-digital conversion. An ADC quantizes each sample to a numeric value; bit depth per channel (commonly 8 bits, giving 24-bit color, with higher-end devices offering more) determines tonal granularity.
• Mechanical transport. Either the sensor/lamp carriage travels the length of the platen (flatbed), the paper is fed past a fixed sensor (sheet-fed/ADF), or the original spins on a drum past a point sensor (drum).
• Assembly and transfer. Successive scan lines are assembled into a full raster and transferred to the host — via USB, SCSI historically, or the network — for further processing.

Architecture

Three sensing architectures dominate, each with a distinct optical and quality profile.

• PMT drum scanners. The original mounts on a rotating drum; a focused point of light and photomultiplier tubes read the image one point at a time. High-end units use three matched PMTs reading red, green, and blue via a dichroic beam splitter. PMTs give very high dynamic range and detail in deep shadows, and drum scanners are cited as reaching very high resolutions (in the range of thousands of PPI, with high-end figures cited up to roughly 24,000 PPI as an illustrative range rather than a fixed spec). They are complex, slow, and expensive — a prepress and archival tool.
• CCD flatbed scanners. A linear CCD array captures a full scan line at once through a mirror-and-lens optical system. CCDs offer good image quality and greater depth of field, useful for scanning slightly non-flat objects such as open books. This is the classic desktop and office architecture.
• CIS scanners. Contact image sensors place the sensor array immediately under the glass, eliminating the mirror-and-lens assembly. This makes the device thinner, lighter, and cheaper, at the cost of shallower depth of field. CIS is common in compact and portable scanners and many low-cost MFPs.

Software architecture separates the device driver from the application, with differing design philosophies. A TWAIN "data source" bundles both device communication and the scanning user interface; SANE cleanly splits a hardware-only backend (driver) from a frontend (UI/application), so the same backend can serve many frontends.

Processing pipeline

A typical scan-to-file pipeline runs through the following stages:

• Acquire raw line data from the sensor and ADC.
• Assemble lines into a raster and apply shading/calibration correction, compensating for lamp and sensor non-uniformity.
• Color and tone processing — color-space handling, gamma, white/black point, and optional color management.
• Geometric correction — deskew, auto-crop, and edge detection; on flatbeds, region detection can separate multiple photos on the platen.
• Enhancement — descreening (to suppress halftone moire from printed originals), dust and scratch handling on film, sharpening, and despeckle.
• Interpretation (optional) — OCR to turn a page image into searchable or editable text, barcode reading, and blank-page removal for document batches.
• Encode and output — write to an image format (TIFF, JPEG, PNG) or a document container (commonly PDF or PDF/A, often with an invisible OCR text layer), then route to file, email, network folder, or cloud.

Advantages

• Converts fragile, bulky, or analog originals into durable, copyable, searchable digital form.
• Enables preservation and access for archives, libraries, and museums, with high-end capture retaining fine detail and wide tonal range.
• Underpins document workflows: searchable PDFs, records management, and paperless offices.
• Is widely accessible: scan capability is now bundled into inexpensive multifunction printers and reachable from phones and any networked client.

Limitations

• Quality is bounded by optics and sensor, not just the advertised dpi; interpolated ("enhanced") resolution does not add real detail. Dynamic range, focus, and color accuracy vary widely by architecture, with drum/PMT generally exceeding CCD, and CCD generally exceeding CIS, in demanding cases.
• Artifacts include moire from printed halftones, dust and scratches on film, and skew or cropping errors on document batches.
• OCR is imperfect, degrading with poor scans, unusual fonts, and complex layouts.
• Speed versus quality trade-offs are pronounced, especially for high-resolution or drum scanning.
• Driver and connectivity fragmentation historically made scanners hard to use across platforms — the very problem TWAIN, SANE, and driverless standards were created to solve.

Relationship to PDF

Scanning and PDF are tightly coupled in document workflows. Scanners commonly output directly to PDF, wrapping page images in a portable container. Two refinements matter. A searchable PDF stores OCR text as an invisible layer aligned to the page image, so the document looks like the scan but is searchable and selectable. PDF/A is the archival PDF subset used for long-term retention of scanned records. These are workflow considerations; format internals belong to a dedicated PDF reference.

Relationship to printers

Scanning is the input counterpart to printing's output, and the two have converged in hardware. Multifunction printers integrate a scanner (usually CCD or CIS, often with an automatic document feeder) with a print engine, enabling copy (scan-then-print) and scan-to-file or scan-to-email. HP explicitly designed the 1987 ScanJet to complement its LaserJet printers for desktop publishing.

On the network side, scanning now rides the same infrastructure as printing: the eSCL scan protocol is the counterpart to Apple's AirPrint, and PWG IPP scanning parallels IPP printing. As Apple's guidance notes, however, print support (AirPrint) does not by itself imply scan support; the device must expose a scanning service.

Relationship to workflows

Scanning is typically the first step in a larger capture workflow: batch document scanning with an ADF, automatic deskew/crop/blank-page removal, OCR, indexing and metadata extraction, and routing into records systems, email, network folders, or cloud storage.

Standards such as TWAIN Direct — a RESTful API for which TWAIN Direct 1.0 was announced in September 2019 — target networked, cross-platform capture for exactly these workflows. Document scanners emphasize throughput, duplex capture, and reliability over the image fidelity prized by photo and archival scanning.

Relationship to standards

• TWAIN (from 1992; TWAIN Working Group): an open API/protocol linking applications and imaging devices via three parts — the application, the Source Manager library, and the device "Source"/data source. It runs on Windows, macOS, and Linux. The name alludes to Kipling's "...never the twain shall meet...," reflecting how hard it once was to connect scanners and PCs; "Technology Without An Interesting Name" is a later backronym.
• SANE (Scanner Access Now Easy): an open API separating hardware backends from UI frontends, and the standard on Linux/Unix. It includes network transport and the scanimage CLI, and its backends include eSCL/AirScan support (sane-escl, sane-airscan).
• ISIS (Image and Scanner Interface Specification): a document-scanning industry standard, cited alongside TWAIN and WIA.
• eSCL / PWG IPP scanning: driverless network scan protocols. Mopria publishes the eSCL specification (the scan side of AirPrint/AirScan), and the PWG defines IPP-based scanning so clients can scan across operating systems without vendor drivers.

Relationship to operating systems

• Windows. Microsoft's imaging stack evolved from the Still Image (STI) architecture to Windows Image Acquisition (WIA), which builds on STI and provides a full device-driver interface plus a system UI. WIA supports network scanners via Web Services for Devices (WSD). Microsoft documents that it now accepts only WIA drivers as in-box scanner/camera drivers and no longer ships TWAIN data sources with Windows, though WIA can offer a TWAIN compatibility layer. TWAIN and ISIS remain widely used at the application level.
• macOS. Apple provides the ImageCaptureCore framework and the Image Capture application for discovering scanners and performing scans; networked all-in-ones are discovered via Bonjour. Apple's own materials distinguish print support (AirPrint) from scan support, which uses the Image Capture architecture.
• Linux/Unix. SANE is the standard interface, with driverless network scanning via eSCL/WSD backends.

Modern relevance

Standalone consumer scanners have largely given way to the scan function inside multifunction printers, smartphone camera "scanning" apps, and high-volume document scanners in offices. The dominant trend is driverless, networked scanning (eSCL/AirScan, PWG IPP scan), which decouples scanning from vendor drivers and OS specifics.

Dedicated high-end scanning — drum/PMT and premium flatbeds — persists for fine-art reproduction, film digitization, and museum or archive preservation where dynamic range and resolution matter. OCR-driven, searchable-PDF capture remains central to records management and the long-running move toward paperless workflows.

Timeline

1. 1843

   Alexander Bain patents an early facsimile process.

2. c. 1848-1851

   Frederick Bakewell improves the design with rotating cylinders and publicly shows a facsimile apparatus at the 1851 Great Exhibition in London.

3. c. 1902

   Arthur Korn's phototelautograph uses a selenium photocell.

4. 1937

   Murray and Morse build an early analog color drum scanner at Eastman Kodak.

5. 1957

   Russell Kirsch's team at NBS scans the first image stored digitally on a computer (SEAC), 176 pixels per side.

6. 1975

   The Autokon 8400, an early flatbed for digital image processing, appears.

7. 1976

   The Kurzweil Reading Machine is unveiled (13 January), combining omni-font OCR, a flatbed scanner, and text-to-speech. ("First omni-font OCR" and "first CCD flatbed" are disputed.)

8. 1984

   The ThunderScan scanning cartridge is released for the Macintosh.

9. 1985

   The Datacopy Model 700, described as the first flatbed scanner for the IBM PC, appears.

10. 1987

    HP introduces the ScanJet flatbed (300 dpi, 4-bit grayscale).

11. 1989

    The HP ScanJet Plus raises bit depth to 8-bit (256 gray levels).

12. 1992

    The TWAIN Working Group is founded and TWAIN 1.0 is released.

13. late 1990s

    The SANE project begins on Unix/Linux.

14. 2000s

    Windows Image Acquisition (WIA) becomes the Windows imaging architecture, building on STI.

15. 2019

    TWAIN Direct 1.0 RESTful API is announced (September).

16. 2010s-2020s

    eSCL/AirScan and PWG IPP driverless network scanning become the mainstream way to scan over the network.