Appendix B

Glossary

OECF, CFA, PSF, gamut, and the rest of the pipeline's vocabulary.

The pipeline’s vocabulary, in one place. Each entry gives the working definition the book uses and the chapter where the idea earns it; where a term of art has a longer story, the chapter is the story.

4:2:0 — video’s ratio notation for chroma subsampling: color-difference planes kept at half resolution in each direction, a quarter of the samples. The mechanism is a 2x2 average. (Chapter 9)

aliasing — what happens when a scene contains frequencies finer than a sampling grid can carry: the detail does not vanish, it returns as confident, wrong, coarser patterns. Moiré is its visible signature. (Chapters 4, 9)

AWB (automatic white balance) — a camera’s estimate of the illuminant from the image alone, descended from the gray-world and white-patch assumptions plus decades of heuristics. (Chapter 3)

Bayer pattern — the 1975 mosaic arrangement: half of all photosites green, a quarter red, a quarter blue, in a 2x2 repeat (RGGB in this book). A bet that the eye wants sharpness in luminance. (Chapters 1, 2)

bilateral filter — an average that discounts neighbors by how different their values are, not just how far away they sit, so smoothing stops at edges. Workhorse of local tone mapping; also a denoiser. (Chapter 7)

black level — the offset a sensor adds before quantizing so that noise around true black survives the trip through unsigned integers. Subtracting it is the pipeline’s first act. (Chapter 2)

CFA (color filter array) — the grid of colored filters over the photosites that makes one sensor produce three-color images; the Bayer pattern is the dominant one. (Chapters 1, 2)

chromaticity — color coordinates with the brightness divided out: what remains of a color when only its hue and saturation matter. White points are stated in it. (Chapters 1, 5)

chrominance / chroma — the color part of an image once luminance has been separated out; what 4:2:0 subsamples and JPEG quantizes hardest. (Chapters 4, 9)

CIE — the international commission that standardizes the measurement of light and color; source of the 1931 standard observer, D65, illuminant A, XYZ, and CIELAB. (Chapters 1, 5)

CIELAB — XYZ bent (mostly by a cube root) into a space where distance roughly matches perceived difference; the home of ΔE. (Chapter 5)

clipping — forcing values into a representable range. At capture it is saturation and destroys information; in the pipeline this book performs it once, at output, as a disclosed decision. (Chapters 0, 2, 7)

color space — a coordinate system for color with a known relationship to human vision. Camera RGB is not one; sRGB and XYZ are. (Chapter 5)

dark frame / flat frame — calibration captures (lens cap on; featureless gray) that reveal hot and dead photosites by where they disagree with their neighbors. (Chapter 2)

ΔE (delta-E) — color difference as distance in CIELAB (this book uses the original 1976 formula). Rule of thumb: 1 is just noticeable, 5 plainly visible, 20 a different color. (Chapter 5)

demosaicing — reconstructing three channels per pixel from a mosaic’s one: prediction from assumptions about images, not recovery of recorded data. (Chapter 4)

display-referred / scene-referred — values meaning “fraction of display brightness” versus values proportional to scene light. The transfer function is the border crossing. (Chapter 7)

distortion (barrel / pincushion) — radial geometric error: straight lines bow outward (barrel) or inward (pincushion). Modeled with one coefficient; corrected by warping. (Chapters 1, 6)

fixed-pattern noise — per-column gain differences that repeat identically in every frame — strictly an error, not noise, and therefore correctable rather than merely reducible. (Chapter 2)

full well — the electron capacity of a photosite. At the ceiling the response is a wall, not a shoulder: linear, then saturated. (Chapter 2)

gamma — the exponent of a display transfer curve (1/2.4 in sRGB’s power segment); by extension, the whole business of nonlinear encoding for display. (Chapter 7)

gamut — the region of colors a space’s primaries can represent; sRGB’s is a triangle the world is not obliged to stay inside. Colors outside map to negative coordinates. (Chapters 5, 10)

gray-world / white-patch — the two classic illuminant estimators: assume the scene averages to gray; assume the brightest thing is white. Each fails on the scene that violates its assumption. (Chapter 3)

hot / dead pixels — photosites that glow in every frame (runaway dark signal) or never respond. Defects are places, not events, which is what makes them correctable. (Chapters 2, 10)

illuminant — the light source’s spectrum. What the camera records is reflectance times illuminant; white balance exists to divide the illuminant back out. (Chapters 1, 3)

ISO — a gain knob applied at readout, after the photons are counted. It brightens signal and noise together; the noise came from the missing light. (Chapter 2)

JFIF — the JPEG File Interchange Format: the header conventions (signature, YCbCr, sampling factors) that make a .jpg file readable everywhere. (Chapter 9)

linear light — values proportional to photon counts, where physics is simple: exposure is multiplication, averaging is legitimate, and displays are unhappy. The pipeline’s native state until tone. (Chapters 0, 2, 7)

luma / luminance — the brightness the eye reads from a color. Luminance is computed on linear values; luma (JPEG’s Y) is the same weighting applied to already-encoded values, as video has always done. (Chapters 7, 9)

LUT (lookup table) — a curve shipped as a table of precomputed values plus interpolation; how pipelines actually carry tone curves. (Chapter 7)

metamerism — two different spectra producing identical responses in some observer. Camera metamerism is the sensor’s version, and no later processing can undo it. (Chapter 5)

middle gray — the conventional 18%-reflectance anchor of photographic tone; the book’s curves pin its rendering in place so changing the look does not change the exposure. (Chapter 7)

moiré — see aliasing; the patterned false detail a regular grid makes from fine regular texture. (Chapter 4)

mosaic — the raw image as captured: one filtered value per photosite, in the CFA’s arrangement. (Chapters 1–4)

OECF (opto-electronic conversion function) — the standard’s name for a capture system’s input-to-output response curve; Chapter 2 measures the sensor’s and finds it linear to the wall.

photosite — one light-collecting well on the sensor: a bucket in the rain. “Pixel” is reserved for the reconstructed image. (Chapter 2)

PSF (point-spread function) — the pattern a single point of light lands as after passing through an optical system; the blur that deconvolution needs named. (Chapters 6, 8)

PSNR (peak signal-to-noise ratio) — image error against a reference on a logarithmic decibel scale; higher is better, +3 dB is a halving of squared error. The book’s currency wherever ground truth exists. (Chapter 4)

quantization — rounding continuous values to a discrete set: to 12 bits at capture, to 8 at output, and — the lossy heart of JPEG — of frequency coefficients under a visibility-weighted table. (Chapters 2, 9)

raw file — a header saying how to read the numbers, then the numbers: the mosaic as the sensor counted it, before any development. (Chapters 1, 10)

read noise — a few electrons of Gaussian fuzz added by the readout electronics whether light arrived or not; why true black must be allowed to average through negative values. (Chapter 2)

ringing — oscillating overshoot along edges produced by negative-lobed kernels and truncated frequency methods: Lanczos ringing, deconvolution moats, JPEG’s ghost ripples are one family. (Chapters 8, 9)

sharpening — raising local contrast so edges read crisper. It adds no detail; it amplifies whatever is there, noise and artifacts included. (Chapter 8)

shot noise — the √N statistics of photon arrival: a property of light, not of the sensor, and the reason no camera takes a noiseless photograph. The only cure is more photons. (Chapter 2)

sRGB — the default display color space: defined primaries, D65 white, and a standard transfer curve. This book’s output space throughout. (Chapters 5, 7)

stop — photography’s factor of two in light. Vision walks in stops, sensors count linearly; most of tone is negotiating between the two. (Chapter 2)

tone mapping — compressing a scene’s brightness range into a display’s, globally with curves or locally with a base/detail split. (Chapter 7)

transfer function — the curve at the display border that spends output levels in perceptual steps rather than linear ones; sRGB’s is the book’s default. (Chapter 7)

unsharp mask — the oldest sharpener: subtract a blurred copy to isolate what changes quickly, add it back scaled. Named for a darkroom sandwich. (Chapter 8)

vignetting — radial light falloff toward the frame’s corners; multiplicative on the way in, divided back out (noise and all) on the way out. (Chapters 1, 6)

von Kries (white balance) — correcting the illuminant by scaling each channel so a neutral surface reads neutral: three multiplications, applied in this book before demosaicing. (Chapter 3)

white point — the chromaticity a system treats as “no color”: D65 for sRGB and this book. (Chapters 1, 5)

XYZ — the CIE’s device-independent color coordinates, built on the standard observer; the space where color claims can be checked. (Chapter 5)

YCbCr — luma plus two color-difference channels; JFIF’s basis, chosen so chroma can be treated roughly. (Chapter 9)

zipper — the demosaicing artifact where alternate rows interpolate an edge differently, turning a straight boundary into sawteeth of false color. (Chapter 4)