Lossy or lossless: which compression should I use for web images?

Use lossy (JPG, WebP, AVIF) for photographs — the human eye barely notices the difference at quality 80–85, and file sizes are 5–20× smaller. Use lossless (PNG, WebP-lossless) for screenshots, UI mockups, logos and anything with sharp edges or text — lossy creates ugly artifacts around boundaries.

Does conversion preserve EXIF metadata (camera, GPS, date)?

It depends on the tool: ImageMagick and FFmpeg copy EXIF by default, while cwebp drops it unless you pass -metadata all . KaijuConverter strips metadata server-side after conversion to protect your privacy — if you need to keep camera or GPS data, use a desktop tool you control.

Lossy or lossless: which compression should I use for web images?

Use lossy (JPG, WebP, AVIF) for photographs — the human eye barely notices the difference at quality 80–85, and file sizes are 5–20× smaller. Use lossless (PNG, WebP-lossless) for screenshots, UI mockups, logos and anything with sharp edges or text — lossy creates ugly artifacts around boundaries.

Does conversion preserve EXIF metadata (camera, GPS, date)?

It depends on the tool: ImageMagick and FFmpeg copy EXIF by default, while cwebp drops it unless you pass -metadata all . KaijuConverter strips metadata server-side after conversion to protect your privacy — if you need to keep camera or GPS data, use a desktop tool you control.

Home
Guides
JPEG Format: Complete Technical Guide to Lossy Compression, DCT, Huffman & Color Subsampling

Guide

JPEG Format: Complete Technical Guide to Lossy Compression, DCT, Huffman & Color Subsampling

PC By Pablo Cirre • Updated May 7, 2026

JPEG: Joint Photographic Experts Group Standard for Lossy Image Compression

JPEG (Joint Photographic Experts Group) is the de facto standard for lossy compression of photographic images. Standardized in 1992, JPEG achieves ~10:1 compression ratios with imperceptible quality loss at moderate quality settings (Q=75–85). JPEG is universally supported—100% of browsers, cameras, and mobile devices natively support it.

JPEG File Structure: SOI, EOI, Markers

JPEG files are built on marker-based segments:

FF D8                      (SOI — Start of Image)
FF E0 00 10 4A 46 49 46 00 (JFIF APP0 marker — defines JPEG variant)
[JFIF identifier and version]
FF DB …                   (DQT — Quantization Table)
FF C0 …                   (SOF0 — Start of Frame, baseline DCT)
FF C4 …                   (DHT — Huffman Table)
FF DA …                   (SOS — Start of Scan)
[compressed image data]
FF D9                      (EOI — End of Image)

Key markers:

Marker	Name	Purpose
FFD8	SOI	Marks beginning of file
FFE0–FFE1	APP0–APP1	Application-specific data (JFIF, EXIF)
FFDB	DQT	Defines quantization tables (quality factor)
FFC0–FFC3	SOF0–SOF3	Start of Frame (specifies MCU dimensions, components)
FFC4	DHT	Huffman table definition
FFDA	SOS	Start of Scan—signals compression data begins
FFD9	EOI	End of Image

JFIF Header: Aspect Ratio & Color Space

FF E0                (APP0 marker)
00 10                (segment length = 16 bytes)
4A 46 49 46 00       ('JFIF' identifier + null terminator)
01 00                (version 1.0)
00                   (units: 0 = no units, 1 = DPI, 2 = DPCM)
00 48                (X density = 72 DPI)
00 48                (Y density = 72 DPI)
00 00                (thumbnail X/Y = none)

Sets aspect ratio and DPI metadata.

Image Encoding: MCU, DCT, Quantization

JPEG divides image into Minimum Coded Units (MCU)—typically 16×16 pixels (2×2 blocks of 8×8):

Step 1: Color Space Conversion (RGB → YCbCr)

Y  = 0.299*R + 0.587*G + 0.114*B   (luminance — perceived brightness)
Cb = -0.169*R - 0.331*G + 0.5*B    (blue chrominance offset)
Cr =  0.5*R - 0.419*G - 0.081*B    (red chrominance offset)

Separates luminance (1x data) from chrominance (2x2 subsampled to 0.25x → 4:2:0 sampling).

Step 2: 8×8 DCT (Discrete Cosine Transform)

DCT transforms 8×8 block of pixel values into frequency coefficients.
Result: [DC coefficient] [low-freq coefficients] [high-freq coefficients]

Concentrates energy in low frequencies—high frequencies are imperceptible to human vision.

Step 3: Quantization (Quality Factor)

Quantized Coefficient = round( DCT Coefficient / Quantization Value )

Quantization values increase for high frequencies (discard imperceptible details).

Example:

Quality = 75 (default):  Quantization value 16 for high freq
Quality = 95 (high):     Quantization value 1 for high freq
Quality = 10 (low):      Quantization value 80 for high freq

Step 4: Huffman Entropy Coding

Variable-length codes: frequent symbols → shorter codes, rare symbols → longer codes
DC coefficient:  -1000 → 5 bits
AC coefficients: 0 (runs) → 3–8 bits

Chroma Subsampling: 4:2:0, 4:2:2, 4:4:4

Human vision is less sensitive to color (chrominance) than brightness (luminance). JPEG exploits this:

Sampling	Y:Cb:Cr	Size reduction	Quality loss	Use Case
4:4:4	1:1:1	None (24 bits/pixel)	Minimal	High-quality (Q≥85)
4:2:2	1:0.5:0.5	33%	Very slight	Typical (Q=75–85)
4:2:0	1:0.25:0.25	50%	Slight	Default (Q≤75)
4:1:1	1:0.25:0.125	62.5%	Noticeable	Extreme compression

Example (4:2:0 chroma subsampling):

Luminance (Y):    16×16 grid (256 values)
Chrominance (Cb): 8×8 grid (64 values)
Chrominance (Cr): 8×8 grid (64 values)
Total:            384 values (vs. 768 uncompressed)
File size:        ~50% of uncompressed

Baseline vs. Progressive JPEG

Baseline JPEG:

SOS marker
[complete MCU data top-to-bottom, left-to-right]
EOI marker

Decoder renders scanlines as data arrives—interlacing not supported.

Progressive JPEG:

SOS marker (scan 1: DC coefficients + low AC)
[partial MCU data]
SOS marker (scan 2: higher AC)
[more detailed data]
[subsequent scans with increasing detail]
EOI marker

First scan shows rough preview, subsequent scans refine. Tradeoff: ~10% larger file but faster perceived load on slow networks.

EXIF & Metadata

EXIF (Exchangeable Image File Format):

FF E1                (APP1 marker)
[EXIF segment with camera metadata]
├── Model: Canon EOS 5D Mark IV
├── Focal Length: 50mm
├── Aperture: f/2.8
├── ISO: 3200
├── Flash: Yes
└── Timestamp: 2024-04-26 12:00:00 UTC

Preserves camera settings, GPS location, copyright info.

Comparison: JPEG vs. Competitors

Metric	JPEG (Q=75)	PNG	WebP	HEIC
File size (photo)	100 KB	300 KB	60 KB	55 KB
File size (graphic)	200 KB	50 KB	40 KB	35 KB
Compression ratio	12:1	3:1	20:1	22:1
Transparency	❌	✅	✅	❌
Progressive	✅	❌	❌	❌
Browser support	100%	100%	97%	25%
EXIF metadata	✅	⚠️ (limited)	❌	✅
Editing reversible	❌ (re-save loses quality)	✅	✅	❌

JPEG remains dominant for cameras and web photos. For web optimization, use WebP. For quality archives, use PNG or TIFF.

Practical JPEG Optimization

Quality settings:

Q=95: Visually lossless (archive)
Q=85: High quality (print, professional)
Q=75: Good quality (default web)
Q=60: Medium quality (thumbnails)
Q=40: Low quality (placeholders)

FFmpeg (baseline JPEG, Q=85):

ffmpeg -i input.jpg -q:v 5 output.jpg

ImageMagick (progressive JPEG):

convert input.jpg -interlace Plane -quality 85 output-progressive.jpg

Python (Pillow):

from PIL import Image
img = Image.open('input.jpg')
img.save('output.jpg', 'JPEG', quality=85, optimize=True)

JPEG's combination of high compression, universal support, and decades of optimization make it the gold standard for photographic images worldwide.

Related conversions

Most teams that read this guide convert images in one of these directions:

Related conversions

Put what you just learned into practice — convert your files now in seconds, free and without registration.

Convert →

Convert →

Convert →

Convert →

Convert →

Convert →

Frequently Asked Questions

JPEG uses lossy compression achieving 10:1 compression ratios (100 KB from 1 MB photo) with imperceptible quality loss at Q=75–85. The DCT transform separates luminance from chrominance, and human vision tolerates chrominance loss. Universal browser/camera support makes JPEG ubiquitous.

JPEG uses com perdas compressão achieving 10:1 compressão ratios (100 KB de 1 MB photo) com imperceptible quality loss at Q=75–85. The DCT transform separates luminance de chrominance, e human vision tolerates chrominance loss. universal browser/camera support makes JPEG ubiquitous.

JPEG uses verlustbehaftet Komprimierung achieving 10:1 Komprimierung ratios (100 KB von 1 MB photo) mit imperceptible quality loss at Q=75–85. The DCT transform separates luminance von chrominance, und human vision tolerates chrominance loss. universell browser/camera support makes JPEG ubiquitous.

JPEG uses con pérdidas compresión achieving 10:1 compresión ratios (100 KB de 1 MB photo) con imperceptible quality loss at Q=75–85. The DCT transform separates luminance de chrominance, y human vision tolerates chrominance loss. universal browser/camera support makes JPEG ubiquitous.

Use <strong>lossy</strong> (JPG, WebP, AVIF) for photographs — the human eye barely notices the difference at quality 80–85, and file sizes are 5–20× smaller. Use <strong>lossless</strong> (PNG, WebP-lossless) for screenshots, UI mockups, logos and anything with sharp edges or text — lossy creates ugly artifacts around boundaries.

Human vision is insensitive to color detail. 4:2:0 subsampling stores color (Cb/Cr) at 1/4 resolution while keeping luminance (Y) full resolution. Result: 50% file size reduction with imperceptible color loss. Chroma subsampling is the primary reason JPEG beats PNG/WebP for photos.

Human vision is insensitive to color detail. 4:2:0 subsampling stores color (Cb/Cr) at 1/4 resolução while keeping luminance (Y) full resolução. Result: 50% tamanho do arquivo reduction com imperceptible color loss. Chroma subsampling is the primary reason JPEG beats PNG/WebP para photos.

Human vision is insensitive to color detail. 4:2:0 subsampling stores color (Cb/Cr) at 1/4 Auflösung while keeping luminance (Y) full Auflösung. Result: 50% Dateigröße reduction mit imperceptible color loss. Chroma subsampling is the primary reason JPEG beats PNG/WebP für photos.

Human vision is insensitive to color detail. 4:2:0 subsampling stores color (Cb/Cr) at 1/4 resolución while keeping luminance (Y) full resolución. Result: 50% tamaño de archivo reduction con imperceptible color loss. Chroma subsampling is the primary reason JPEG beats PNG/WebP para photos.

For new projects in 2026, WebP is supported by all modern browsers (95%+ of traffic) and saves 25–35% over JPG at the same visual quality. AVIF is even more efficient (40–50% smaller) but encoding is slower and Safari support is recent. Use WebP as the default and AVIF as the progressive enhancement via <code><picture></code> with JPG fallback.

Progressive JPEG encodes multiple passes—first showing low-resolution preview, then refining. Slightly larger file (~10%), but appears faster on slow networks. Use for hero images on web. For thumbnails or small images, baseline JPEG is fine.

Progressive JPEG encodes multiple passes—first showing baixa-resolution preview, then refining. Slightly larger arquivo (~10%), mas appears faster on slow networks. usar para hero images on web. para thumbnails ou small images, baseline JPEG is fine.

Progressive JPEG encodes multiple passes—first showing niedrig-resolution preview, then refining. Slightly larger Datei (~10%), aber appears faster on slow networks. verwenden für hero images on web. für thumbnails oder small images, baseline JPEG is fine.

Progressive JPEG encodes multiple passes—first showing baja-resolution preview, then refining. Slightly larger archivo (~10%), pero appears faster on slow networks. usar para hero images on web. para thumbnails o small images, baseline JPEG is fine.

It depends on the tool: ImageMagick and FFmpeg copy EXIF by default, while cwebp drops it unless you pass <code>-metadata all</code>. KaijuConverter strips metadata server-side after conversion to protect your privacy — if you need to keep camera or GPS data, use a desktop tool you control.

No. Each JPEG save applies quantization—re-saving compounds losses. JPEG is destructive. Always edit from lossless source (RAW, PNG, TIFF), save once as JPEG. Never JPEG→edit→JPEG.

No. Each JPEG salvar applies quantization—re-saving compounds losses. JPEG is destructive. Always editar de sem perdas source (RAW, PNG, TIFF), salvar once as JPEG. Never JPEG→edit→JPEG.

No. Each JPEG speichern applies quantization—re-saving compounds losses. JPEG is destructive. Always bearbeiten von verlustfrei source (RAW, PNG, TIFF), speichern once as JPEG. Never JPEG→edit→JPEG.

No. Each JPEG guardar applies quantization—re-saving compounds losses. JPEG is destructive. Always editar de sin pérdidas source (RAW, PNG, TIFF), guardar once as JPEG. Never JPEG→edit→JPEG.

Three common causes: (1) ICC color profiles dropped during conversion shift colors slightly; (2) chroma subsampling (4:2:0) reduces color accuracy in lossy JPG/WebP; (3) the new format may not support all features of the original (e.g. PNG → JPG drops transparency to white). For pixel-perfect results stick to lossless formats and preserve color profiles explicitly.