Video Coding

Video coding represents digital video in a compressed form for storage or transmission.

Video Processing Stages

1. Digitise the source video.
2. Process colour, contrast, and brightness as required.
3. Encode and compress the resulting video.

Initial compression during capture reduces data volume but affects final quality.

Video Characteristics

Frame rates used in the lecture:

• Cinema
  24 frames per second.
• PAL television
  25 frames per second.

Uncompressed Video Rate

The lecture’s 768 × 576 video uses 24 bits per pixel at 25 frames per second.

$768 \times 576 \times 24 \times 25 = 265{,}420{,}800\ \text{bps}$

Approximately 265.4 Mbps before transmission overhead.

Compression Trade-Offs

Redundancy is repeated or predictable information that can be represented more compactly.

• Reducing dimensions removes spatial detail.
• Reducing frame rate removes temporal detail.
• Lossy coding can discard information less noticeable to human perception.
• Strong compression can produce blockiness or other visible distortion.
• Rapid motion reduces similarity between consecutive frames.

General-Purpose Compression

General-purpose lossless methods exploit repeated values or unequal symbol probabilities.

Run-Length Encoding

Run-length encoding (RLE) replaces consecutive occurrences of a symbol with the symbol and its run length.

• Effective for long runs of identical values.
• Suits uniform image regions, graphics, and cartoons.

Examples:

• 111110000003355 becomes (1, 5), (0, 6), (3, 2), (5, 2).

Relative Encoding

Relative encoding stores an initial value followed by differences between consecutive values.

Small differences can require fewer bits than the original values.

Huffman Coding

Huffman coding assigns variable-length prefix codes to symbols according to their frequencies.

• Frequent symbols receive shorter codes.
• Effective when a small number of symbols dominate.

Arithmetic Coding

Arithmetic coding represents a symbol sequence by progressively narrowing an interval according to symbol probabilities.

• Can encode a sequence more efficiently than symbol-by-symbol Huffman coding.
• Requires more complex processing.

Intraframe Compression

Intraframe compression exploits spatial redundancy within an individual image or video frame.

Subsampling

Spatial subsampling reduces the number of pixels used to represent an image.

• Pixel selection
  Retains a subset of the original pixels.
• Averaging
  Represents a region using an average pixel value.

Averaging generally gives better quality than direct pixel selection.

Coarse Quantisation

Coarse quantisation reduces the number of possible values used to represent each pixel.

Preserves the number of pixels but reduces their precision.

Vector Quantisation

Vector quantisation represents a block of input values using a reference to a similar pattern in a shared codebook.

Example adapted from November 2023, Question 4.3.

• The input block is (10, 10, 10, 10).
• The closest codebook entry is (12, 12, 12, 12).
• Decoding its index reconstructs every sample with an error of +2.
• Repeating the input block repeats the same distortion.

Repetition alone does not cause a large error. A poorly matched codebook can produce persistent distortion even for highly repetitive input.

Transform Coding

Transform coding represents image blocks using coefficients in a transform domain.

DCT coding uses 8 × 8 blocks, followed by quantisation, run-length coding, and entropy coding.

Interframe Compression

Interframe compression exploits temporal redundancy across a sequence of video frames.

Frame Subsampling

Frame subsampling transmits selected frames and reduces the effective frame rate.

Difference Coding

Difference coding compares a frame with its predecessor and updates changed pixels.

Block-Based Difference Coding

Block-based difference coding compares non-overlapping blocks with their counterparts in a reference frame.

Only significantly changed blocks are updated.

Block-Based Motion Compensation

Motion compensation predicts a current block using a similar region in a reference frame.

• Motion vector
  Specifies the displacement to the reference region.
• Residual
  Represents the difference between the predicted block and the current block.

Encoding Selection Examples

The choice depends on motion, repeated information, and the detail the application must preserve. These techniques can be combined within a codec.