# Digital Capture and the Input-Output Chain

Photography I · Photographic Language and the Taking of the Image · https://tryals.app/learn/photography-i/digital-capture-and-the-input-output-chain

## From Photons to a File

A sensor is a grid of **photosites**. Each photosite counts photons and returns a number.

Three steps convert this count into an image:
- **A colour filter array.** Photosites cannot detect colour. Most sensors use a **Bayer array**, created for Kodak by Bryce Bayer. Half the filters are green, while red and blue each take a quarter.
- **Demosaicing.** Every final pixel needs three colours. Two-thirds of colour values are interpolated from neighbours.
- **Quantisation.** The analogue count converts to a digital integer.

| Bit depth | Levels per channel |
|---|---|
| 8-bit | 256 |
| 12-bit | 4096 |
| 14-bit | 16384 |

### Linear Encoding

Raw sensor data is **linear**. Doubling the light doubles the recorded value.

In a 14-bit File, the brightest stop takes **8192** levels. The next takes 4096. Exposing to the right preserves far more data.

### Raw vs JPEG

| Feature | Raw | JPEG |
|---|---|---|
| Bit depth | 12 or 14 | 8 |
| Colour | One value per site | Three values per pixel |
| Processing | Unbaked metadata | Baked in irreversibly |

### Print Output

A standard hand-held Print requires about 300 ppi.

A 6000 by 4000 pixel file prints at 20 by 13.3 inches at 300 ppi. A 16 by 24 inch print needs more pixels.

> **Common pitfall:** treating megapixels as a measure of quality. A 24 megapixel phone file differs hugely from a 24 megapixel full-frame file. Pixel count sets maximum print size, not image fidelity.

## Practice questions

9 of this lesson's 14 practice questions, with answers. The full set is in the app.

### 1. Sort each step by where it sits in the input-output chain.

**Answer:**

- Input: Lens projects image, Photosite quantisation
- Processing: Demosaicing pixel data, Tonal and colour edits
- Output: ICC profile print mapping, File resampled to 300 ppi

**Why:** The three stages are worth keeping separate because they fail differently. Input losses are irrecoverable, processing losses are recoverable if the original is kept, and output losses are recoverable simply by printing again. Most panic about image quality is about the only stage where nothing is lost permanently.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/sort-each-step-by-where-it-sits-in-the-input-output-chain

### 2. Half the photosites under a Bayer array carry a green filter and each photosite measures only one colour, yet the finished file gives red, green and blue for every pixel. How is that possible?

A. Each photosite rapidly cycles through red, green and blue measurements during exposure
B. The file records only luminance at capture, deriving three colour channels from white balance
C. The missing two thirds of colour data is interpolated from neighbours during demosaicing
D. Green photosites record enough luminance to mathematically deduce exact red and blue values

**Answer:** C. The missing two thirds of colour data is interpolated from neighbours during demosaicing

**Why:** By interpolation. One measurement per photosite and three values per pixel means two thirds of the colour in every digital photograph is calculated from neighbours rather than measured. Demosaicing does this very well, and where it shows is fine coloured detail close to the sampling limit.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/half-the-photosites-under-a-bayer-array-carry-a-green-filter-and-each

### 3. A Bayer colour filter array carries twice as many green filters as red or blue.

**Answer:** True

**Why:** **True.** Half the filters are green, a quarter red and a quarter blue. The asymmetry follows the eye: most of our perception of fine detail comes from the green part of the spectrum, so sampling it twice as densely buys resolution where the viewer will actually notice it.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/a-bayer-colour-filter-array-carries-twice-as-many-green-filters-as

### 4. A 24-megapixel sensor records 14 bits at every photosite. How large is one uncompressed raw file, in megabytes, taking a megabyte as one million bytes?

**Answer:** 42 (within ±1)

**Why:** **42 MB.** 24 million photosites at 14 bits is 336 million bits, and dividing by 8 gives 42 million bytes. Note that this is one value per photosite, not three: a demosaiced 8-bit TIFF of the same picture is 72 MB and a 16-bit one is 144 MB.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/a-24-megapixel-sensor-records-14-bits-at-every-photosite-how-large

### 5. Complete the account of digital capture.

**Answer:** The mosaic of coloured filters over a sensor is the **Bayer** array, in which half the filters are green. Reconstructing three colours for every pixel from one measurement each is called **demosaicing**. Raw data is **linear**, so twice the light gives twice the number. A print held in the hand is normally made at about **300** pixels per inch.

**Why:** The third blank is the one with consequences. Linear encoding means the levels are distributed by light rather than by appearance, so the shadows, where the eye is most sensitive to banding, are described by the fewest numbers, which is the entire argument for exposing to the right.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/complete-the-account-of-digital-capture

### 6. Which are true of bit depth and linear encoding?

A. A higher bit depth increases the dynamic range the sensor can capture
B. In a 14-bit file spanning fourteen stops, the brightest stop takes half the available levels
C. Editing a file in 8 bits leaves less headroom for large tonal moves than editing in 16
D. An 8-bit channel holds 256 levels and a 14-bit channel holds 16384

**Answer:** B. In a 14-bit file spanning fourteen stops, the brightest stop takes half the available levels; C. Editing a file in 8 bits leaves less headroom for large tonal moves than editing in 16; D. An 8-bit channel holds 256 levels and a 14-bit channel holds 16384

**Why:** The fourth conflates two things. Dynamic range is set by the sensor's full-well capacity against its noise floor; bit depth only decides how finely that range is sliced. More bits over the same range means smoother gradients, not more shadow detail than the sensor could see.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/which-are-true-of-bit-depth-and-linear-encoding

### 7. A raw file is an image that simply has not been compressed yet.

**Answer:** False

**Why:** **False.** A raw file is a set of single-colour measurements plus metadata, not an image: it has not been demosaiced, white balanced or tone mapped. That is why two converters produce visibly different pictures from the identical file, and why "the raw file" is not a picture anyone has ever seen.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/a-raw-file-is-an-image-that-simply-has-not-been-compressed-yet

### 8. A 6000 by 4000 pixel file is printed at 300 pixels per inch. How wide is the print, in inches?

**Answer:** 20 (within ±1.5)

**Why:** **20 inches**, so the print is 20 by 13.3 inches, a little over 50 centimetres on the long side. Doubling that print size means halving the resolution to 150 ppi, which is entirely acceptable for a picture viewed from two metres and visibly soft in the hand.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/a-6000-by-4000-pixel-file-is-printed-at-300-pixels-per-inch-how-wide

### 9. Sensor data is recorded linearly, whereas human vision perceives brightness logarithmically. Why does this physical characteristic make underexposing a raw capture so damaging to image fidelity?

A. Darker stops receive far fewer quantisation steps
B. Shadows drop below the analogue detection threshold
C. Linear values cannot be interpolated by demosaicing
D. Eight-bit conversion truncates darker stops entirely

**Answer:** A. Darker stops receive far fewer quantisation steps

**Why:** Because light values halve with each descending stop, the lowest stops share only a tiny fraction of the total digital levels. Noise dominates shadow recovery, whereas highlight headroom retains thousands of discrete tones.

Page: https://tryals.app/practice/photography-i/digital-capture-and-the-input-output-chain/sensor-data-is-recorded-linearly-whereas-human-vision-perceives