What an unprocessed photo looks like

Original Article ↗ Hacker News Discussion ↗

Choice of Example and Aesthetics

  • Several readers liked the explanation but felt the Christmas-tree scene (harsh mixed LED light, drab subject) makes it hard to judge what a “good” result should look like.
  • Others counter that the point is precisely to show how unappealing very basic processing is, and that real cameras apply much more sophisticated pipelines.

What “Unprocessed” Actually Means

  • Strong agreement that a truly unprocessed photo you can look at doesn’t exist: raw sensor output is just per‑pixel voltages, usually in a Bayer mosaic, not RGB.
  • The first dark “linear” images are already processed (ADC, rescaling to 0–255, mapping to sRGB); they’re just processed differently from the camera JPEG.
  • Some argue that gamma and simple per‑pixel transfer functions are just encoding choices (like decompression), not “editing,” while others see all these steps as legitimate processing.

Gamma, Tone Mapping, and Displays

  • Large subthread on gamma correction vs linear light:
    • One side: if monitors and files had enough bit depth, we could stay linear end‑to‑end.
    • Others: human perception is nonlinear, so you must introduce a nonlinearity somewhere; gamma/log encoding also optimizes bit usage.
  • Distinction drawn between gamma encoding and tone mapping for compressing dynamic range (e.g., avoiding blinding sunsets).

Bayer Pattern, Green Dominance, and Luminance

  • Extended discussion of why Bayer is RGGB: green carries most perceived luminance/detail, matches human sensitivity and many YUV/YCbCr luminance formulas.
  • Clarifications about alternative sensor layouts (X‑Trans, Foveon), monochrome sensors, and how RAW formats are already somewhat processed.
  • Side tangents into grayscale conversion coefficients, historic TV standards, and color vision/colour blindness.

Real vs Fake, Edits, and AI

  • Big thread on what counts as a “fake” image:
    • One camp: all photos are interpretive; only intent to deceive makes them fake.
    • Others: there’s a meaningful spectrum—from global tone/contrast tweaks and demosaicing to object-level edits, generative fill, and scene alteration.
  • Journalistic norms are cited: global, uniform adjustments OK; adding/removing elements or per-object AI enhancement generally not.
  • Examples discussed: moon “enhancement” on phones, skin smoothing, removing unwanted objects, and oversaturated travel/foliage photos.

Phones, DSLRs, Noise, and Sharpening

  • Complaints about aggressive in‑camera denoising and sharpening, especially on phones and cheap IP/dash cams (plastic “painted” look, missing details, license plates).
  • Some note that RAW on phones or separate apps can avoid this, but mass‑market defaults optimize for flattering, punchy images, not fidelity.

Broader Takeaways and Pointers

  • Many appreciate how the post demystifies the image pipeline and underscores that both film and digital photography are layers of signal processing and aesthetic choice.
  • Several recommend image-processing textbooks, astrophotography workflows, open RAW tools (dcraw/libraw), and other deep‑dive blog posts and videos for further exploration.