Restore an old photo

Behind the results

How archival scanning resolution (PPI), bit depth, and enlargement math work

When preserving a family photo collection—whether shoeboxes of 1950s deckle-edge drugstore snapshots, trays of 35mm Kodachrome slides, or 19th-century mounted Cabinet Cards—the two most common mistakes are underscanning tiny film transparencies (such as scanning 35mm slides at 600 PPI, which produces a blurry postage-stamp image) and overscanning large paper prints (such as scanning an 8×10 inch silver-gelatin print at 4,800 PPI, which bloats a single file past 10 gigabytes while capturing only paper fiber texture).

To eliminate guesswork, federal preservation agencies—specifically the Federal Agencies Digital Guidelines Initiative (FADGI, 3rd Edition, May 2023) and the National Archives and Records Administration (NARA 36 CFR § 1236.50)—define objective optical sampling targets based on the physical size and resolving power of the original medium.

1. Optical PPI vs. Printer DPI and enlargement magnification

Scanner manufacturers frequently advertise interpolated resolutions like "9,600 × 9,600 dpi," but archival quality depends strictly on true Optical Pixels Per Inch (PPI_scan) sampled by the scanner’s sensor across the physical width (W_in) and height (H_in) of the film frame or paper print:

Captured Pixel Dimensions:
  W_px = round(W_in × PPI_scan)
  H_px = round(H_in × PPI_scan)
  Megapixels (MP) = (W_px × H_px) / 1,000,000

Maximum Uninterpolated Print Size at Target Print Density (PPI_print):
  W_print_max = W_px / PPI_print = W_in × (PPI_scan / PPI_print)
  H_print_max = H_px / PPI_print = H_in × (PPI_scan / PPI_print)

Required Optical Scan PPI for a Target Enlargement (W_target × H_target):
  Magnification M = max(W_target / W_in, H_target / H_in)
  PPI_required = ceil(M × PPI_print)

Worked example (35mm slide vs. 4×6 print): A standard 35mm slide or negative strip frame measures just 1.417 × 0.945 inches (36 × 24 mm). If you want to print an 8 × 10 inch framed portrait at 300 PPI without artificial pixel upscaling, the required magnification along the short axis is M = 8.0 / 0.945 = 8.47×, requiring ceil(8.47 × 300) = 2,540 PPI minimum—which is why FADGI 3-Star specifies 3,200 PPI (yielding 4,535 × 3,024 px = 13.71 MP) and FADGI 4-Star specifies 4,000 PPI (5,669 × 3,780 px = 21.43 MP). By contrast, a 4 × 6 inch paper snapshot already contains its full silver-halide detail by 600 PPI (2,400 × 3,600 px = 8.64 MP), allowing a crisp 8 × 12 inch print at 300 PPI. Read our complete walkthrough in Best PPI Resolution for Scanning Old Family Photos and Slides.

2. Uncompressed bit-depth and archival storage equations

An 8-bit channel stores 2^8 = 256 discrete tonal steps from pure black (0) to pure white (255). When you stretch the histogram of a faded 1920s silver print or a magenta-shifted 1970s Kodacolor print in 8-bit mode, those 256 steps spread apart, leaving empty "comb gaps" that appear as harsh banding in skies and skin tones. Scanning in 16 bits per channel (48-bit RGB or 16-bit Grayscale) provides 2^16 = 65,536 tonal gradations per channel (281.47 trillion RGB color combinations). The exact byte footprint of an uncompressed raster master is:

Bytes_uncompressed = W_px × H_px × C × (b / 8) + 32,768 (TIFF IFD + ICC profile header)
  where C = 3 for RGB Color (1 for Grayscale), and b ∈ [8, 16] bits per channel.

Decimal Megabytes (MB, storage drive rating) = Bytes / 10^6
Binary Mebibytes (MiB, OS file manager rating) = Bytes / 2^20 (1,048,576)

Applying lossless LZW or Deflate compression inside a TIFF container (κ ≈ 0.68) or saving a 16-bit lossless PNG (κ ≈ 0.62) shrinks storage by roughly 30–38% while keeping every single bit mathematically identical to the uncompressed scan. For details on master vs. access derivatives, see TIFF vs. PNG vs. JPEG: Archival File Formats and Bit-Depth Math.

Reference table: FADGI 2023 & NARA 36 CFR § 1236.50 optical PPI standards across 10 formats

Home movie & oral-history audio digitization bitrates

Magnetic videotapes (VHS, VHS-C, Video8, Hi8, MiniDV) suffer from binder hydrolysis ("sticky-shed syndrome") and head-clogging oxide shedding after 25–40 years. Converting hours of analog or DV tape into digital storage follows the linear bitrate equation:

Video Storage (GB) = (Bitrate_Mbps × 3,600 seconds/hour × Hours) / 8,000 bits-per-byte-ratio

Learn how to prevent interlaced comb artifacts (converting 29.97i fields to 59.94p progressive frames) and calculate multi-drive CISA 3-2-1 backup arrays in our guides on Digitizing VHS, Hi8, MiniDV, and 8mm Home Movies and ISO 8601 File Naming and the 3-2-1 Backup Rule.

Reading RGB histograms & protecting real faces during AI photo restoration

In Part 2 above, our client-side canvas inspector evaluates every pixel using the ITU-R BT.709 luminance equation (Y = 0.2126 R + 0.7152 G + 0.0722 B), which weights green highest and blue lowest to match human retinal sensitivity. When a 1960s–1980s chromogenic print turns pink or brick-red, its cyan dye layer (which absorbs red light) has decayed in dark storage, causing the red channel mean (μ_R) to surge 20–45 levels above the green channel (μ_G).

When using modern generative AI tools to repair cracked emulsion or severe dye fade, unconstrained prompts ("restore and enhance this old photo") frequently trigger face-hallucination models that replace a grandparent’s true eye spacing, nose bridge, and age lines with a generic, airbrushed stranger. Our Part 3 compiler injects explicit Facial Geometry Lock negative constraints to encourage conservative repairs. These instructions cannot guarantee that a generative model will preserve identity; always compare its output with your original. Explore the full methodology in Reading Photo Histograms to Fix Faded Magenta and Yellow Prints and AI Photo Restoration Prompts That Preserve Real Facial Features, or switch to our Tombstone Birth-Date, 1752 Calendar, US Census Window & Cousin DNA Calculator to date the ancestors in your newly scanned portraits.