Flatbed vs Sheet Fed Scanner for Legal Files and OCR

A flatbed scanner holds the page still on a sheet of glass and moves a sensor beneath it. A sheet fed scanner pulls each page past a fixed sensor. For legal files, feed sound loose pages and put fragile, bound or irreplaceable originals on the glass. OCR reads either, provided the image is clean.

The choice is less about the machine than about the paper. A feeder is quick and handles the paper roughly. Glass is slow and gentle. Each also leaves its own kind of defect in the image, and knowing which defect belongs to which device tells you where a bad page came from and how to get a better one.

What is the difference between a flatbed and a sheet fed scanner?

On a flatbed the paper stays put and the sensor travels. On a sheet fed scanner the sensor stays put and the paper travels, drawn through by rollers from a tray called an automatic document feeder. That one difference decides speed, the risk to the original, and which faults show up in the file.

The two designs reach an office as three devices:

• A flatbed. A lid, a pane of glass (the platen) and nothing else. One face at a time, placed by hand.
• A desktop sheet fed scanner. A tray, a set of rollers and a slot. No glass to lay a page on.
• A multifunction copier. Both in one body: a feeder on top that lifts to reveal a flatbed.

A few terms come up in manuals and in conversations with whoever services the machine. The pick roller takes the top sheet off the stack. The separation pad or retard roller holds the remaining sheets back so only one goes through. A carrier sheet is a clear plastic sleeve that lets a small or delicate item ride through a feeder inside a full-size envelope.

The last row has its own post. Whether a feeder captures the back of each sheet, and how backs get lost when it does not, is explained under duplex scanning and two-sided originals.

Which legal originals should never go through a document feeder?

Keep an original off the feeder when damage to it could not be undone or when its physical state may matter later. That covers bound volumes, brittle or torn paper, anything with a seal, ribbon or attachment, and any original whose staples, folds or fastenings you have not been cleared to disturb.

A feeder needs loose, flat, reasonably strong sheets. Getting an original into that state means changing it: pulling staples, cutting a binding, peeling off a note. On a working copy that is housekeeping. On an original it may be an alteration.

The glass is the place for these:

• Original wills, codicils and other testamentary papers. Do not remove a staple or fastening from one to make it feedable until the responsible lawyer says so. Whether a disturbed fastening has any consequence is a question for your jurisdiction, and the cautious course is to leave it alone and scan each page on the glass.
• Bound books. Corporate minute books, ledgers, registers, diaries and notebooks.
• Instruments with seals or ribbons. Wax, foil and ribbon will not survive rollers, and neither may the rollers.
• Old, brittle or tissue-thin paper. Carbon copies on onionskin, aged newsprint and paper that cracks along its folds.
• Torn or taped pages. Tape catches in the paper path. A tear becomes two pieces.
• Photographs and glossy prints. Rollers can mark the surface.
• Thermal paper. Till receipts and old fax rolls curl, and the coating can darken under heat or pressure.
• Small items. Cheques, receipts, business cards, postal slips and identity cards, unless you have a carrier sheet.
• Pages carrying sticky notes or attachments. Scan once with the note in place and once with it moved, both on the glass.
• Originals held as evidence. If staple holes, fold lines, indentations or the order of the sheets could ever be in issue, ask before you change anything, and record the condition before scanning.

A sort before scanning

1. Fan the stack and pull out anything bound, stapled, taped, glossy, undersized or fragile.
2. Ask whether any of the remaining sheets is an original that must stay exactly as it is. If so, pull it too.
3. Feed what is left.
4. Scan the pulled items on the glass, and put a note in the file saying which pages came from which device, so the next reader understands why some pages look different.

Double feeds, skew and streaks: faults a sheet feeder introduces

A sheet feeder adds faults that come from moving paper: two sheets drawn through as one, pages that enter at an angle, and vertical lines from dirt sitting in front of the fixed sensor. Each leaves a recognizable trace in the file, and each can change or remove text without any warning on the screen.

Double feeds

A double feed happens when two sheets go through stuck together. The scanner sees only the top one, so the file is a page short and nothing in the file says so. Static, damp paper, a worn separation pad, and sheets that were stapled or folded together all encourage it. Some feeders have a sensor that stops the job when it detects two thicknesses; check whether yours does and whether it is switched on.

The only dependable check is counting. Count the sheets before you feed, compare with the page count afterwards, and follow the printed page numbers through the file.

Skew

A sheet that enters the rollers at an angle is recorded at that angle, and a sheet that twists partway through is recorded with lines that fan out. Loose paper guides, mixed paper sizes in one stack and dog-eared corners all cause it. A folded corner does double damage: it tilts the sheet and hides whatever is printed under the fold.

Streaks

On a feeder the sensor looks through a narrow strip of glass, and every page passes over the same strip. A fleck of correction fluid, toner or adhesive on that strip blocks the same few pixels on every line of every page. The result is a thin vertical line from top to bottom, in the same place on each page.

That position is how you tell a streak from something on the paper. A line printed on the original moves when the paper shifts. A streak does not.

A streak matters to OCR more than its width suggests. It can pass through a column of figures and turn a 1 into something else, read as a letter l or a vertical bar in the middle of a word, or pass for the ruling of a table.

Pull one deed out and scan it on the glass before a long feeder run. Then look up every party, plus any citation or notice printed in the margin. A streak through that citation hides a fact you may look up later. The glass holds the sheet still while you capture those lines.

Glass, lid and book spine: faults a flatbed introduces

A flatbed adds faults that come from placing paper by hand: dirt on the glass that prints onto every image, a lid that does not close over thick originals, pages that curve away at a book's spine, and human slips such as a skipped or repeated page. The original stays safe even when the image is not.

The glass

Dust, a hair, a fingerprint or a smear of adhesive on the platen appears in every image at the same spot. Unlike a feeder streak it is a speck or a smudge and not a line, because the sensor travels past it once. A speck beside a figure can read as a decimal point or a comma. Clean the glass with whatever the manufacturer's manual specifies before a long job, and again when the same mark turns up on two pages running.

The lid

The lid presses the page flat and blocks room light. When a thick book or a stapled bundle holds it open, three things follow: the page lifts off the glass and the print softens, room light gets in and greys the background, and the area around the page records as solid black. A black surround is harmless to the text but confuses automatic cropping and inflates the file.

The spine

A bound volume opened face down does not lie flat. Near the gutter the page rises from the glass, the lines of text bend, and a shadow falls across the inner margin. Words there may be dark, narrow and out of focus all at once. Pressing harder risks the binding. For what recognition makes of bent lines and how to handle them, see the section on curved pages from bound volumes.

The size of the glass

A flatbed cannot record what hangs over the edge of its platen. Measure the glass before you scan legal-size paper on it. If the glass is shorter than the sheet, the foot of every page is missing from the image, so check your own pages for signatures, initials and page numbers in that strip.

The operator

Every page is a separate act, so the errors are human ones: a page skipped, a page scanned twice, a sheet turned over twice so the same face is captured again, a page laid upside down or at an angle. A simple discipline guards against them. Move each sheet from an "unscanned" pile on the left to a "done" pile on the right, face down, and say the printed page number as you place it.

Does the type of scanner change what OCR can read?

The type of scanner changes what OCR can read only through the image it produces. Recognition software receives pixels and no record of the device that made them. What differs between a flatbed and a feeder is the kind of defect each adds, and recognition suffers from the defect, not from the device.

Recognition software has no way to know whether rollers or a moving sensor produced a page. So the useful question about a poor result is which fault is in the image:

• Sloping lines and a short page count point to the feeder.
• A vertical line in the same place on every page points to the feeder's glass strip.
• Soft, bent text near one margin points to a bound original on a flatbed.
• The same speck on every page points to the platen.
• A missing foot on every page points to a sheet longer than the glass.

Resolution matters on both devices in the same way. A page scanned at too few dots per inch is hard to read whichever machine made it, and the guide to the best resolution for OCR explains how to choose the setting.

Testing the two on your own page

If your office has both devices and you are unsure which to use for a batch, test with a page from the batch.

1. Pick the hardest page: the smallest print, the faintest copy, or the one with figures in columns.
2. Scan it through the feeder and on the glass, with the same resolution and color mode.
3. Combine the two images into one two-page PDF, feeder first.
4. Run recognition on the file.
5. Compare the text of the two pages against the paper, reading names, dates and amounts character by character.

Schema Law accepts PDF, PNG and JPEG uploads of up to 20 MB, and the free plan allows one document a day of up to three pages, so the two-page test file fits. The home page says recognition is strongest on clean print and that nothing marks a passage as uncertain, which is why step 5 is a reading against the paper and not a glance at the output. The pricing page lists the page allowances for larger jobs.

Office copier, desktop feeder or flatbed: matching the device to the file

Match the device to the condition of the paper first and the size of the job second. Loose, sound, uniform sheets go through a feeder. Anything bound, fragile, undersized or untouchable goes on glass. A mixed file gets both, with a note saying which pages were scanned how.

Two points about the copier. Its feeder and its flatbed are separate scanners sharing a control panel, so a fault on one says nothing about the other. And its scan settings are whatever was last saved as the default, which may have been chosen for small email attachments and not for recognition. Check resolution and color mode on the panel before a job that matters.

When you send paper out to be scanned, the same vocabulary lets you give instructions that can be checked: which items are to stay off the feeder, whether fastenings may be removed, and what the vendor should report back, such as sheet counts per folder and a list of pages scanned on glass.