Compression: lossy, lossless and RLE

What you need to be able to do

Why compress?

Compression reduces the file size. On its own that earns little; every mark comes from the consequence in the context given:

Lossy and lossless compression

Run-length encoding (RLE)

RLE on text

1. Read from the start. Count how many times the same character appears in a row (consecutively).
2. Write the count, then the character.
3. Start counting again at the next different character.

Note that the final “AA” is a new run: RLE only joins identical values that are next to each other.

RLE on a bitmap

Each pixel's colour code is the value. A row of 20 pixels, W = white, B = black:

Decoding

Read each pair as “this many of this value”: 3F2 = F2 F2 F2. In hex-coded images the count comes first and the colour is the last two hex digits: “11D” = one pixel of colour 1D.

When RLE does not help

RLE stores a value and a count for every run. If the data has few repeated runs, every run has length 1, so each value gains a count and the file gets bigger: Red Green Blue → Red 1 Green 1 Blue 1 (June 2023). Photographs, with many slightly different colours next to each other, often compress badly with RLE; simple graphics with large areas of one colour compress well.

How each kind of file is compressed

Vector graphics compress much less than bitmaps: the drawing list has little redundant data to remove (November 2021 scheme).

Choosing lossy or lossless: case by case

Mark-scheme focus

Lossless compression of a text file; why lossy must not be used (4 marks)

Credited

• (iii) Run-length encoding can be used B1; identifies groups of repeated characters … replaces them with one copy of the character and the number of times it occurs B1
• (iv) None of the original data can be lost / deleted B1; the text file would be corrupted // cannot be opened B1

Not credited

• (iii) describing image compression (pixels) for a text file
• (iv) one point only, written twice in different words

Why

The June 2022 report: most said the file would be corrupted but missed the second mark. Two marks = two distinct points: no data may be lost, AND what happens if it is.

From: 9618/13/M/J/22 Q1(a)(iii), (iv)

Why RLE may not reduce the size of a bitmap, with an example (3 marks)

Credited

• RLE stores a colour and the number of times it occurs consecutively B1
• An image may not have many sequences of the same colour B1
• It would need to store each colour and then the count 1, which adds data B1 (max 2 for the explanation)
• Example: Red-Green-Blue would become Red 1 Green 1 Blue 1 B1

Not credited

• An example with characters (“ABC → 1A1B1C”) when the question is about a bitmap
• “RLE doesn't work on photos” with no reason

Why

The example must be about colours / pixels: the June 2023 report noted examples using characters instead of colours.

From: 9618/11/M/J/23 Q1(c)(ii)

Why lossy compression is often suitable for a bitmap image (2 marks)

Credited

• The change may not be noticeable // data removed is usually not noticed by the human eye B1
• … for example, changes in shade / detail B1
• It produces a larger decrease in file size compared to lossless B1 (max 2)

Not credited

• “It makes the file smaller” (true of all compression)
• “The quality stays the same” (it doesn't - it just isn't noticed)

Why

The reason lossy suits images is human perception plus a bigger reduction; “smaller” alone doesn't separate lossy from lossless.

From: 9618/13/O/N/24 Q6(b)(iii)

Why compress a sound file for email; how lossless compression works on sound (4 marks)

Credited

• (i) Reduces the file size B1; faster to transmit / download // original file too large for email storage / attachment B1
• (ii) e.g. RLE: where consecutive sounds are the same, record the value and the number of times it repeats B1 B1 // reduce the amplitude range to only the range used, so fewer bits per sample // record the changes instead of the actual sounds

Not credited

• (i) “Easier to transmit” (the report wanted speed)
• (ii) generic lossless descriptions with text or image examples

Why

(ii) must be about sound: the report said answers were generic and seldom referred to the sound file.

From: 9618/11/O/N/21 Q7(b)

Worked examples

Example 1 · RLE on one row of pixels (Paper 1, 2 marks) ★ Easy

1 Read it with me what is actually happening?

Ten identical pixels in a row: the perfect case for RLE. The second mark is for applying it to this row.

• this row of pixelsuse the row: 10 × B
• losslessRLE

2 What is it asking for? the target, part by part

• (d)name + apply2

3 What do we know? everything the question gives us

• BBBBBBBBBB

4 Plan the route work backwards from what we want

FindKey ideaAlready have itWork out firstTrap

• Compress
  • MethodRLE
  • Applycolour B, count 10

What would a generic answer miss?

The report: answers that didn't explain how the method applies to the row given lost the second mark.

5 Now solve it have a go on paper first, then open the worked solution

1. Use run-length encoding B1
2. Record the colour Blue and the number of times it occurs, 10 B1

6 Check it does the answer make sense?

10 colour codes become one colour + one count ✓.

Example 2 · Completing an RLE table, then text and lossless (Paper 1, 4 marks) ★★ Medium

1 Read it with me what is actually happening?

The first row shows the format: count, then the 2-digit hex colour. Row 2 is decoding; row 3 is encoding.

• first row completedcopy its format exactly
• each colour = hex valuecolour codes are 2 hex digits; anything before them is the count

2 What is it asking for? the target, part by part

• (c)(i)two cells2
• (c)(ii)why lossless for text2

3 What do we know? everything the question gives us

• 2AB 2FF 11D 167decode
• 32 32 80 81 81encode

4 Plan the route work backwards from what we want

FindKey ideaAlready have itWork out firstTrap

• Row 2decode
  • Split2|AB, 2|FF, 1|1D, 1|67
  • Trapwriting FF once
• Row 3encode
  • Runstwo 32s, one 80, two 81s
  • Trap80 instead of 180

How is “11D” split?

Count 1, colour 1D: the colour is always the last two hex digits.

5 Now solve it have a go on paper first, then open the worked solution

1. (i) Row 2: AB AB FF FF 1D 67 B1
2. Row 3: 232 180 281 B1
3. (ii) All the data is required // no data can be lost B1 … otherwise the text file will be corrupted / not make sense B1

6 Check it does the answer make sense?

Re-encode row 2 to check: AB AB → 2AB, FF FF → 2FF, 1D → 11D, 67 → 167 ✓. The report saw “only one FF” and “80 instead of 180”.

Example 3 · Live streaming a concert: choose and justify (Paper 1, 3 marks) ★★ Medium

1 Read it with me what is actually happening?

Three marks, all for the justification. Real-time streaming = the data must keep arriving as fast as it is watched.

• real-timeno time to wait: buffering ruins it
• subscribersmany viewers, many devices
• justifyreasons linked to streaming

2 What is it asking for? the target, part by part

• (a)choice + justification3

3 What do we know? everything the question gives us

4 Plan the route work backwards from what we want

FindKey ideaAlready have itWork out firstTrap

• Lossy?
  • Sizemuch smaller → less bandwidth
  • Viewerloss not noticed; less buffering
  • Devicesphones don't need full resolution

Could lossless be justified?

Yes, the scheme also credits it (e.g. viewers don't want any loss; high-bandwidth devices). Lossy is the stronger case.

5 Now solve it have a go on paper first, then open the worked solution

1. Lossy (ticked)
2. Loss of quality will not be noticed B1
3. Needs to be viewed in real time, so less bandwidth needed if the file size is smaller B1
4. Smaller file sizes reduce buffering so the video plays more smoothly // viewers on different devices may not need high resolution B1

6 Check it does the answer make sense?

Every point links the choice to real-time streaming ✓.

Summary

• Compression reduces file size → less storage, faster transfer, less bandwidth / buffering: always tie it to the context.
• Lossless: no data lost, original re-created exactly (text, code). Lossy: data permanently removed, much smaller (photos, audio, video, streaming).
• RLE: runs of consecutive identical values → value + count. Great for long runs, makes data without runs bigger.
• Image lossy: lower depth / palette / resolution. Sound lossy: lower rate / resolution, remove inaudible sounds.

Can I do this?

Quiz

Concept checks

Answer each one in your head first, then open it.

Exam-style practice

New questions, not used above. Do them on paper, then mark your work against the official mark scheme.