TIFF vs. PNG vs. JPEG: Archival File Formats and Bit-Depth Math

Every scanner driver asks for a file format (TIFF, PNG, or JPEG) and bit depth (8-bit Grayscale, 16-bit Grayscale, 24-bit RGB, or 48-bit RGB). Choosing 8-bit JPEG for an irreplaceable shoebox of 19th-century cabinet cards or faded 1970s Ektachrome slides permanently discards up to 99.99% of raw tonal gradations and bakes 8 × 8 block artifacts into paper grain. Conversely, saving every healthy modern snapshot as an uncompressed 48-bit TIFF can consume six times more storage than necessary. You can calculate exact per-file and whole-collection storage footprints across every format and bit depth using the Photo & Slide Digitization Planner.

Exact Uncompressed Byte-Size Mathematics

A digital raster image is a two-dimensional rectangular matrix of width W_px pixels and height H_px pixels. Each pixel stores C color channels (C = 1 for monochrome grayscale; C = 3 for Red, Green, and Blue additive color), and each channel is quantized into b bits (BitsPerChannel, either 8 bits or 16 bits). Because 1 byte = 8 bits, the exact uncompressed image payload (Bytes_raw, before container headers or embedded ICC profiles of 4 KB to 64 KB) is:

Total Pixels:
  N_px = W_px × H_px

Bytes per Pixel (Bpp):
  Bpp = C × (b ÷ 8)

Raw Uncompressed Payload (Bytes):
  Bytes_raw = W_px × H_px × C × (b ÷ 8) = N_px × Bpp

Decimal Megabytes (SI / Storage Drive Manufacturers):
  MB = Bytes_raw ÷ 1,000,000

Binary Mebibytes (Operating System File Explorers, 2^20 bytes):
  MiB = Bytes_raw ÷ 1,048,576

Comparing the Four Standard Archival Bit Modes

Scanner software labels bit depth either by bits per channel (8-bit vs. 16-bit) or by total bits per pixel (24-bit vs. 48-bit). The table below reconciles both naming conventions and shows the exact byte multiplier per pixel:

Scanner Mode Label	Channels (C)	Bits / Channel (b)	Total Bits / Pixel	Bytes / Pixel (Bpp)	Tones / Channel (2^b)	Total Possible Values ((2^b)^C)
8-bit Grayscale	1 (Luminance)	8 bits	8 bits	1 Byte/px	256 steps	256 gray shades
16-bit Grayscale	1 (Luminance)	16 bits	16 bits	2 Bytes/px	65,536 steps	65,536 gray shades
24-bit RGB Color	3 (R, G, B)	8 bits	24 bits	3 Bytes/px	256 steps	16,777,216 (16.78 million)
48-bit RGB Color	3 (R, G, B)	16 bits	48 bits	6 Bytes/px	65,536 steps	281,474,976,710,656 (281.47 trillion)

Worked Arithmetic: 4×6 Print (600 PPI) vs. 35mm Slide (4,000 PPI)

Let us apply the uncompressed byte equation to the two primary archival targets derived in Best PPI Resolution for Scanning Old Family Photos and Slides:

1. Standard 4 × 6 in Print Scanned at 600 PPI (2,400 × 3,600 px = 8,640,000 px):

   • 8-bit Grayscale (1 B/px): 8,640,000 × 1 = 8,640,000 Bytes (8.64 MB / 8.24 MiB).
   • 16-bit Grayscale (2 B/px): 8,640,000 × 2 = 17,280,000 Bytes (17.28 MB / 16.48 MiB).
   • 24-bit RGB (3 B/px): 8,640,000 × 3 = 25,920,000 Bytes (25.92 MB / 24.72 MiB).
   • 48-bit RGB (6 B/px): 8,640,000 × 6 = 51,840,000 Bytes (51.84 MB / 49.44 MiB).

2. Mounted 35mm Slide (24 × 36 mm) Scanned at 4,000 PPI (3,780 × 5,669 px = 21,428,820 px):

   • 24-bit RGB (3 B/px): 21,428,820 × 3 = 64,286,460 Bytes (64.29 MB / 61.31 MiB).
   • 48-bit RGB (6 B/px): 21,428,820 × 6 = 128,572,920 Bytes (128.57 MB / 122.62 MiB).

Across a carousel box of 500 Kodachrome slides scanned at 4,000 PPI in 48-bit uncompressed TIFF, raw storage totals 500 × 128,572,920 Bytes = 64,286,460,000 Bytes, or 64.29 GB (59.87 GiB, where 1 GiB = 1,024 MiB) for a single working copy.

---

Why 16-Bit Per Channel Prevents Histogram Combing on Faded Photos

Human vision under ideal viewing conditions distinguishes roughly 200 to 250 discrete luminance steps between pure black and paper white, which is why monitors and sRGB JPEGs use 8 bits per channel (2^8 = 256 integer values from 0 to 255). Why, then, do FADGI and NARA guidelines recommend capturing faded, stained, or high-density photographic originals in 16 bits per channel (65,536 values from 0 to 65,535)?

The Quantization Math of Stretching a Faded Channel

Suppose you scan a faded 1972 chromogenic print or underexposed 35mm slide whose surviving Blue channel dye spans only 15.625% of the scanner’s dynamic range. Look at what happens when you expand that narrow range back to full contrast (0 to 255) using the Levels transformation detailed in Reading Photo Histograms to Fix Faded Magenta and Yellow Prints:

Case A: Scanned in 8-bit per channel (256 total steps)
  Captured active steps: 256 × 0.15625 = 40 discrete tonal steps (bins 90 to 129)
  Stretched across 256 output bins:
    Occupied output bins = 40
    Empty zero-pixel gaps = 256 - 40 = 216 missing bins (84.4% comb gaps!)

Case B: Scanned in 16-bit per channel (65,536 total steps)
  Captured active steps: 65,536 × 0.15625 = 10,240 discrete tonal steps
  Downsampled to 256 output bins after tonal stretching:
    Available input steps per final 8-bit bin = 10,240 ÷ 256 = 40.0 sub-steps per bin
    Empty zero-pixel gaps = 0 (100% smooth continuous histogram!)

When you stretch a 40-step 8-bit capture across [0, 255], adjacent input values jump by 255 ÷ 40 ≈ 6.4 levels at a time. On the image histogram, those missing integers appear as vertical white gaps resembling the teeth of a hair comb (histogram combing). On the photograph itself, smooth sky gradients, coat lapels, and skin tones break apart into stair-stepped contour bands (posterization). Capturing in 16-bit per channel (48-bit RGB or 16-bit Grayscale) provides 256 internal sub-steps inside every 8-bit step, allowing aggressive black/white point expansion with zero comb gaps.

---

Archival Format Showdown: Uncompressed TIFF vs. Lossless TIFF vs. PNG vs. JPEG

Not all file containers handle bit depth, compression, and provenance metadata equally. Below is how the four major raster formats compare under Library of Congress (fdd000022) and NARA (36 CFR § 1236.50) preservation criteria.

Feature / Criterion	Uncompressed TIFF (.tif)	Lossless LZW / Deflate TIFF (.tif)	Lossless PNG (.png)	Lossy JPEG (.jpg)
Compression Algorithm	None (raw byte array)	Lossless LZW or ZIP/Deflate	Lossless Deflate + Paeth filter	Lossy 8 × 8 DCT + Quantization
Typical Size vs. Raw	100% (1.00×)	55% to 75% (0.55× to 0.75×)	50% to 70% (0.50× to 0.70×)	5% to 15% (0.05× to 0.15× of 24-bit)
16-Bit/Channel (48-bit)	Yes (native standard)	Yes (Deflate preferred for 16-bit)	Yes (supported in PNG spec)	No (standard baseline JPEG is 8-bit)
Metadata Support	Full TIFF Tags, EXIF, IPTC, XMP	Full TIFF Tags, EXIF, IPTC, XMP	Limited (chunks often stripped)	EXIF, IPTC, XMP, ICC
Re-Save Generation Loss	Zero (100% bit-identical)	Zero (100% bit-identical)	Zero (100% bit-identical)	Cumulative loss on every re-save
FADGI / NARA Role	Primary Preservation Master	Approved Preservation Master	Web / intermediate graphic	Access Derivative Only

1. Uncompressed Baseline TIFF 6.0 (.tif)

Tracked by the Library of Congress as format fdd000022, Uncompressed Baseline TIFF 6.0 stores pixel values as a contiguous byte array accompanied by an Image File Directory (IFD) header. Even if a storage drive suffers a bad sector decades from now, an uncompressed TIFF remains readable across all undamaged rows—whereas a corrupted byte inside a compressed stream can break the rest of the image.

2. Lossless LZW and Deflate (ZIP) Compressed TIFF

Both LZW and Deflate / ZIP (approved under NARA 36 CFR § 1236.50) are mathematically 100% lossless: when decompressed into memory, every pixel is bit-for-bit identical to an uncompressed TIFF (MSE = 0). On 8-bit grayscale or 24-bit RGB print scans, LZW or Deflate shrinks file size by 25% to 45%. On 16-bit (48-bit RGB) film scans, sensor shot noise in the lowest bits reduces LZW efficiency unless paired with horizontal differencing (Predictor 2) and Deflate.

3. Portable Network Graphics (.png)

While PNG supports 24-bit and 48-bit lossless RGB using Deflate compression and row-adaptive filtering, it was engineered as a web graphic format rather than a museum master. Many cataloging tools fail to parse XMP/IPTC archival provenance fields inside PNG chunks, and single-threaded PNG encoding at high compression (level 9) is slower during batch scanning than TIFF.

4. Why JPEG (.jpg) Is Never an Archival Master

The JPEG format achieves 10:1 to 20:1 file-size reduction through three irreversible steps:

1. Chroma Subsampling (4:2:0): Standard JPEG converts RGB into YCbCr and averages each 2 × 2 block of color channels (Cb and Cr) into a single value—discarding 75% of spatial color resolution.
2. 8 × 8 Discrete Cosine Transform (DCT) Quantization: Every 8 × 8 block of 64 pixels is transformed into spatial frequencies and divided by a quantization matrix set by the Quality slider (1–100), rounding high-frequency coefficients to zero.
3. Generation Loss on Re-Saving: Every time you open a JPEG, rotate it, adjust brightness, and click Save, the 8 × 8 DCT quantization runs again, compounding mosquito noise around handwritten ink captions and blocky grids across faces.

---

The NARA and Library of Congress Two-Tier Archival Strategy

Professional archives combine preservation permanence with easy family sharing using a Two-Tier File Architecture:

• Tier 1 — Archival Preservation Master (_master.tif): Scan once at FADGI 3-Star or 4-Star optical resolution into an Uncompressed or Lossless Deflate TIFF. Use 48-bit RGB (Adobe RGB 1998) for faded color prints, slides, and negatives requiring histogram restoration, or 24-bit RGB / 16-bit Grayscale (Gray Gamma 2.2) for stable monochrome prints. Store masters read-only following ISO 8601 File Naming and the 3-2-1 Backup Rule for Family Archives.
• Tier 2 — Access & Sharing Derivative (_access.jpg): After tonal balancing in 16-bit mode, export an 8-bit JPEG (Quality 90–92, 4:4:4 chroma, sRGB) for emailing relatives, online family trees, or digital picture frames. For VHS cassettes, Hi8 tapes, or Super 8 reels, apply the same master-plus-derivative architecture in Digitizing VHS, Hi8, MiniDV, and 8mm Home Movies: Bitrate and Storage.