JPEG Image

Compression pipeline

JPEG encoding runs through four stages. First, the image converts from RGB to YCbCr color space, separating luminance (Y) from chrominance (Cb, Cr). Human vision resolves brightness at higher spatial frequency than color, so chrominance channels are typically downsampled 2:1 horizontally (4:2:2) or both horizontally and vertically (4:2:0), halving or quartering the color data before any math begins.

Second, each channel splits into 8×8 pixel blocks. Third, each block passes through the Discrete Cosine Transform, converting spatial pixel values into 64 frequency coefficients. The DC coefficient represents the block's average brightness; the remaining 63 AC coefficients capture increasingly fine detail.

Fourth, a quantization matrix divides each coefficient and rounds to the nearest integer. This is where data is destroyed. High-frequency coefficients get divided by large values, pushing them to zero. The quality slider in any image editor controls how aggressive this quantization is. At quality 85, a typical photograph retains visually indistinguishable output. Below 50, block artifacts emerge.

The 8×8 block boundary problem

JPEG's DCT operates on 8×8 pixel blocks independently, creating visible grid artifacts under heavy compression — the block size was chosen in 1992 as a balance between efficiency and consumer hardware processing power. Each block encodes with no awareness of its neighbors. At low quality, adjacent blocks quantize differently, producing a tiled checkerboard visible on smooth gradients. Modern decoders in Chrome and Safari apply post-processing deblocking filters, but the artifact is baked into the bitstream.

Performance benchmarks

Decode speed is JPEG's hidden advantage. libjpeg-turbo, the standard decoder on Linux and Android, processes a 12 MP image in 25–40 ms using SIMD-accelerated IDCT. WebP decode runs 1.3–2× slower for equivalent dimensions. AVIF decode is 3–5× slower. On low-power mobile devices, that gap matters for scroll performance.

Progressive vs baseline

Baseline JPEG encodes blocks left-to-right, top-to-bottom in a single scan. Progressive JPEG stores multiple scans of increasing detail — the first scan renders a blurry full-frame preview in roughly 15% of the file, then subsequent scans sharpen it. Progressive files are 2–5% smaller than baseline at the same quality because the multi-scan layout allows more efficient Huffman coding of AC coefficients.

When to use JPEG vs alternatives

JPEG remains the best choice for photographs distributed to unknown clients — email attachments, CMS uploads, legacy device support. WebP offers 25–30% smaller files at equal quality and supports transparency, making it superior for web delivery when format negotiation is available. AVIF compresses 40–50% smaller but decodes slowly and lacks support in older Safari versions. PNG is the right choice only when lossless output or alpha transparency is required.

Limitations

No transparency. No animation (MJPEG wraps frames in AVI or MOV, not in JPEG itself). Maximum dimension is 65,535 pixels per side. Color depth caps at 8 bits per channel in standard JPEG; higher-bit workflows require JPEG 2000 or JPEG XL. And JPEG's patent-free status, settled after the Forgent Networks dispute in 2006, means no licensing cost — a practical advantage that helped cement its ubiquity.