Multimedia Data Compression & Coding: Techniques, Trade-offs, and Real-World Impact

Why Compress Multimedia?

Multimedia files (images, audio, video) are data-rich but inefficient. A single 1080p video can exceed 1GB per minute if uncompressed. Compression shrinks files to:

• Save storage (e.g., NTC’s 5G network needs to transmit compressed video calls).
• Reduce bandwidth (e.g., WhatsApp’s video calls use H.264 compression to avoid buffering).
• Speed up transfers (e.g., Daraz’s product images are JPEG-compressed to load faster).

Compression works by exploiting redundancy (repeated patterns) and perceptual limits (what humans can’t notice).

1. Lossless vs. Lossy Compression: The Core Trade-off

graph TD
    A["Multimedia Data"] --> B["Lossless Compression"]
    A --> C["Lossy Compression"]
    B --> D["No quality loss\n(Archives, docs)"]
    C --> E["Minimal quality loss\n(Videos, music)"]
    D --> F["Examples: ZIP, PNG-8\n(No data loss)"]
    E --> G["Examples: JPEG, MP3\n(Small file, slight artifacts)"]

Worked Example: PNG vs. JPEG

• A PNG of a logo (text + flat colors) compresses to 1.2MB (lossless).
• The same logo as a JPEG compresses to 80KB (lossy), but you’d notice color banding in high-zoom views.

2. Entropy Coding: The Mathematical Foundation

Entropy coding exploits probability—frequent symbols (e.g., spaces in text) get shorter codes.

Key Methods

1. Huffman Coding
   • Assigns shorter bits to more frequent symbols.
   • Example: In English text, ‘e’ (12%) gets 0, ‘z’ (0.1%) gets 111111.

graph TD
    A["Symbol"] --> B["Frequency Count"]
    B --> C["Assign Variable-Length Codes"]
    C --> D["Huffman Tree Example"]
    D --> E["Example: 'e' → 0, 'z' → 111111"]
    E --> F["Prefix-Free Property"]

2. Arithmetic Coding
   • Encodes entire sequences as a single number (e.g., 0.101101 for "hello").
   • Better for low-frequency symbols (e.g., rare words in a book).

Real-World Use:

• WhatsApp’s text messages use Huffman coding to compress repeated words like “ok” or “yes.”
• ZIP files combine Huffman + LZ77 (see below) for lossless archives.

3. Predictive Coding: Exploiting Spatial/Temporal Redundancy

Predictive coding estimates repeated patterns (e.g., smooth gradients in images) and stores only the difference (error).

A. Run-Length Encoding (RLE)

• Stores sequences of identical values as (value, count).
• Example: A black-and-white bar chart’s long white stripe might encode as (255, 1000).

B. Lempel-Ziv-Welch (LZW)

• Used in GIF/PNG and TIFF.
• Replaces repeated substrings with a single code (e.g., “the” → 12345).
• Limit: Patent expired in 2003, but still used in Nepal’s eSewa transaction IDs (compressed for storage).

4. Transform Coding: The Secret to JPEG/MP3

Transform coding converts data into a domain where redundancy is easier to remove.

Discrete Cosine Transform (DCT)

• Breaks an image into 8×8 blocks and transforms them into frequency components (low/high).
• Key Insight: Humans can’t see high-frequency noise (e.g., jagged edges).
• Process:
  1. Split image into 8×8 blocks.
  2. Apply DCT → frequency spectrum.
  3. Quantize (round) high frequencies → lossy compression.
  4. Encode remaining coefficients with Huffman.

graph TD
    A["Original Image"] --> B["8x8 Pixel Block"]
    B --> C["DCT Transform → Frequency Spectrum"]
    C --> D["Quantization (Discard High Frequencies)"]
    D --> E["Huffman Encoding → JPEG File"]
    E --> F["Result: Smaller File, Slight Artifacts"]

Worked Example: JPEG Compression Trace

1. Start with a 512×512 RGB image (15MB uncompressed).
2. Split into 8464 blocks (512/8 × 512/8).
3. DCT reveals 90% of data is low-frequency (smooth colors).
4. Quantize high frequencies → file shrinks to 2MB (JPEG).
5. Huffman encodes → final size: 1.5MB.

Real-World Impact:

• Daraz’s product images use JPEG to load in <2s on slow 3G.
• NEPSE stock charts use DCT-compressed PNGs to reduce server load.

5. Hybrid Coding: Combining Techniques

Most real-world compression (e.g., JPEG, MP3, H.264) uses multiple methods:

1. DCT (for frequency analysis).
2. Quantization (lossy reduction).
3. Entropy coding (Huffman/arithmetic).

Example: MP3 (Audio Compression)

• Steps:
  1. Sampling: Convert analog audio to 44.1kHz PCM (16-bit samples).
  2. DCT: Split into frequency bands.
  3. Masking: Remove inaudible frequencies (e.g., >16kHz for humans).
  4. Bitrate reduction: Store only critical bands (e.g., 128kbps MP3 vs. 1.4MB WAV).

graph TD
    A["Original Audio"] --> B["Sampling: 44.1kHz PCM"]
    B --> C["FFT/DCT → Frequency Bands"]
    C --> D["Psychoacoustic Masking"]
    D --> E["Quantize & Encode → MP3"]
    E --> F["128kbps vs. 1.4MB WAV"]

Why MP3?

• A 3-minute song goes from 100MB (WAV) → 3MB (MP3).
• Ncell’s music app uses MP3 to save storage for offline playlists.

6. Video Compression: MPEG/H.264

Video compression reuses frames to avoid storing duplicates.

Key Techniques

1. Intraframe (I-frames): Full JPEG-like compression (e.g., keyframes).
2. Predictive (P-frames): Store differences from previous frame.
3. Bidirectional (B-frames): Interpolate between two I/P frames.

graph TD
    A["I-Frame"] --> B["Full JPEG Compression"]
    B --> C["P-Frame"]
    C --> D["Motion Vectors + Differences"]
    D --> E["B-Frame"]
    E --> F["Interpolation from I/P Frames"]
    F --> G["Efficient for Video"]

Real-World Use:

• Pathao’s ride-sharing app uses H.264 to stream driver location updates smoothly.
• NTC’s live TV compresses to 5Mbps (vs. 200Mbps uncompressed).

7. Challenges in Multimedia Compression

Example: JPEG Artifacts

• Problem: High compression (e.g., 10%) creates blockiness.
• Fix: JPEG 2000 uses wavelet transforms for smoother edges.

In the Real World

1. Khalti’s Payment Gateway

   • Idea: Lossless compression for transaction logs (PNG-8 for icons, ZIP for metadata).
   • Why? Ensures audit trails are 100% accurate (no data loss).

2. Google Photos

   • Idea: Hybrid DCT + neural networks (Google’s "Smart Compression").
   • How? Uses AI to discard unnoticeable details (e.g., tiny text in photos).
   • Result: 15:1 compression with near-lossless quality.

3. Ncell’s 4K Video Calls

   • Idea: H.265 (HEVC) compression (50% smaller than H.264).
   • Why? 4K video would need 10Gbps without compression—now it’s 2–3Mbps.

Exam Tip

Focus on these 3 areas for full marks:

1. Compare Lossless vs. Lossy (table + real examples like PNG vs. JPEG).
2. Trace JPEG/H.264 (show DCT → quantization → Huffman steps).
3. Hybrid Coding (explain how MP3/JPEG combine DCT + entropy coding).

Common Pitfalls:

• ❌ Forgetting why high frequencies are discarded (perceptual limits).
• ❌ Mixing up RLE (for runs) and LZW (for substrings).
• ❌ Not linking to real apps (e.g., "Ncell uses H.265 for 4K calls").

Sample Answer Structure:

1. Define (e.g., "Entropy coding reduces file size by assigning shorter codes to frequent symbols").
2. Explain with a diagram (e.g., Huffman tree for "hello").
3. Give a real example (e.g., "WhatsApp uses Huffman for text compression").
4. Compare (e.g., "Lossless: ZIP; Lossy: JPEG").

Final Visual Recap