Why JPEG DCT Struggles With Synthetic Graphics

Synthetic graphics featuring hard edges, solid colors, and high-contrast text compress poorly under standard JPEG compression because the Discrete Cosine Transform (DCT) relies on continuous mathematical waves to represent image data. Designed originally for natural photography with soft transitions, the DCT algorithm approximates sharp pixel changes by combining varying cosine frequencies. When JPEG quantizes and discards the high-frequency data needed to render these abrupt boundaries, it introduces visible distortion known as ringing artifacts, mosquito noise, and edge blurring.

The Mathematics Behind the Discrete Cosine Transform

The baseline JPEG standard breaks an image into \(8 \times 8\) pixel blocks and converts the spatial pixel values of each block into frequency domain coefficients using the Discrete Cosine Transform. This mathematical operation expresses the block as a weighted sum of 64 distinct cosine waveforms ranging from low to high spatial frequencies.

In natural photographs, neighboring pixels usually change gradually, meaning most visual energy is concentrated in the low-frequency components. Synthetic imagery—such as screenshots, vector icons, diagrams, and typography—consists of discontinuous step functions where adjacent pixels transition instantaneously between polar values (for example, black text on a white canvas).

The Frequency Penalty of Sharp Transitions

According to Fourier analysis, an abrupt step function cannot be accurately represented by a small number of continuous waves. Recreating an instantaneous boundary requires an extensive, theoretically infinite set of high-frequency cosine harmonics.

When a synthetic graphic is broken into \(8 \times 8\) DCT blocks:

• High-Frequency Saturation: Rather than isolating the energy into a few low-frequency coefficients, a sharp edge scatters energy across nearly all 64 frequency coefficients in the block.
• Loss of Compression Efficiency: Because so many high-frequency coefficients are populated with large values, the block loses its mathematical compressibility. The image file size increases disproportionately relative to its visual complexity.

Quantization and Ringing Artifacts

The primary lossy step in JPEG compression is quantization, which divides the DCT coefficients by predetermined values and rounds the results to zero. Because the human eye is relatively insensitive to subtle high-frequency variations in natural scenes, JPEG quantization tables aggressively discard high-frequency data.

When applied to synthetic graphics, this creates critical visual flaws:

1. Gibbs Phenomenon and Ringing: Stripping the high-frequency harmonics prevents the cosine waves from cleanly canceling each other out at the boundary of a sharp edge. The remaining truncated frequencies produce an oscillating ripple effect around high-contrast borders, commonly called "ringing" or "mosquito noise."
2. Block Boundary Mismatches: Because DCT processes pixels in isolated \(8 \times 8\) grids, a continuous sharp vector line passing through multiple blocks is approximated slightly differently in each one. This accentuates the borders of the grid, creating glaring blockiness along sharp lines.
3. Color Bleeding (Chroma Subsampling): JPEG often pairs DCT with chroma subsampling (such as 4:2:0), which halves the color resolution horizontally and vertically. While largely invisible on real-world textures, subsampling blurs vibrant colored text and hard-edged graphic elements before the DCT transformation even occurs.

Format Suitability

Because the DCT is inherently built around wave-based smoothing, formats utilizing palette-based indexing, run-length encoding, or dictionary-based compression (such as PNG, SVG, or lossless WebP) are technically superior for synthetic graphics. These formats handle uniform color fields and sharp, single-pixel transitions natively without attempting to approximate discrete boundaries through continuous frequencies.