JPEG encoding, chroma subsampling, and MozJPEG tuning

JPEG remains the most universally compatible lossy format. In production pipelines, tuning Quality alone leaves major gains on the table: chroma subsampling controls whether coloured text edges look blurry, while progressive scans (Interlace) and Huffman/trellis quantization typically cut file size by 5%–15% at the same visual quality.

Cross-binding JPEG parameter reference

Setting	govips JpegExportParams	sharp .jpeg(options)	pyvips / vips CLI jpegsave	Production guidance
Quality factor	Quality (default 80/75)	quality (default 80)	Q=82 / --Q 82	78–85 for web photography
Chroma subsampling	SubsampleMode	chromaSubsampling	subsample_mode	auto (4:2:0) for photos; off (4:4:4) for sharp red/blue text or UI badges
Progressive scan	Interlace: true	progressive: true	interlace=True	Recommended for images above 10 KB
Huffman optimization	OptimizeCoding: true	optimiseCoding: true (default)	optimize_coding=True	Losslessly shrinks output; keep enabled
Trellis & MozJPEG tuning	TrellisQuant, OvershootDeringing, OptimizeScans, QuantTable	mozjpeg: true (enables preset bundle)	trellis_quant, overshoot_deringing, optimize_scans, quant_table=3	Reduces bitrate when linked against MozJPEG at higher CPU cost
Metadata stripping	StripMetadata: true	Stripped by default (.keepMetadata() to retain)	keep="none" (or strip=True)	Removes embedded camera EXIF thumbnails

Production JPEG export examples

The snippets below configure Quality = 82, progressive scan, Huffman coding optimization, and MozJPEG quantization options across all four interfaces:

govips

package main

import (
    "fmt"
    "os"

    "github.com/davidbyttow/govips/v2/vips"
)

func main() {
    if err := vips.Startup(nil); err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
    defer vips.Shutdown()

    image, err := vips.NewImageFromFile("input.jpg")
    if err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
    defer image.Close()

    // JPEG does not support alpha: flatten onto white if the input may have transparency
    if image.HasAlpha() {
        if err := image.Flatten(&vips.Color{R: 255, G: 255, B: 255}); err != nil {
            fmt.Fprintln(os.Stderr, err)
            os.Exit(1)
        }
    }

    params := vips.NewJpegExportParams()
    params.Quality = 82
    params.Interlace = true
    params.OptimizeCoding = true
    params.SubsampleMode = vips.VipsForeignSubsampleAuto
    params.StripMetadata = true

    // Advanced quantization flags (effective when libvips is linked with MozJPEG):
    params.TrellisQuant = true
    params.OvershootDeringing = true
    params.OptimizeScans = true
    params.QuantTable = 3

    data, _, err := image.ExportJpeg(params)
    if err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
    if err := os.WriteFile("output.jpg", data, 0o644); err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
}

sharp

import sharp from 'sharp';

// mozjpeg: true enables trellisQuantisation, overshootDeringing, optimiseScans, progressive, and quantisationTable: 3
await sharp('input.jpg')
  .flatten({ background: '#ffffff' })
  .jpeg({
    quality: 82,
    progressive: true,
    chromaSubsampling: '4:2:0',
    mozjpeg: true,
  })
  .toFile('output.jpg');

pyvips

import pyvips

image = pyvips.Image.new_from_file("input.jpg")
if image.hasalpha():
    image = image.flatten(background=[255, 255, 255])

image.write_to_file(
    "output.jpg",
    Q=82,
    interlace=True,
    optimize_coding=True,
    subsample_mode="auto",
    trellis_quant=True,
    overshoot_deringing=True,
    optimize_scans=True,
    quant_table=3,
    keep="icc",
)

vips CLI

#!/usr/bin/env bash
set -euo pipefail

vips jpegsave input.jpg output.jpg \
  --Q 82 \
  --interlace \
  --optimize-coding \
  --subsample-mode auto \
  --trellis-quant \
  --overshoot-deringing \
  --optimize-scans \
  --quant-table 3 \
  --keep icc

Chroma subsampling: when to switch from 4:2:0 to 4:4:4

Human vision is more sensitive to luminance (Y) detail than chrominance (Cb/Cr). Under default SubsampleMode = Auto:

• When Q < 90, libvips uses 4:2:0 chroma subsampling (each 2 × 2 pixel block shares one chroma sample), halving chroma data;
• When Q >= 90, libvips automatically switches to 4:4:4 (no subsampling).

If you export images containing fine red/blue typography, promotional badges, or UI screenshots at Q = 82, disable chroma subsampling explicitly (vips.VipsForeignSubsampleOff / chromaSubsampling: '4:4:4' / subsample_mode="off") to prevent colour bleeding around text edges.

Memory cost of progressive JPEG

Progressive JPEG (Interlace: true / progressive: true) requires buffering the full plane of DCT coefficients in memory during encoding and decoding. It is well worth enabling for standard web dimensions (such as under 2000 × 2000), but on multi-megapixel archival images it raises peak memory significantly — keep baseline encoding (Interlace: false) for very large dimensions.

Sources

• libvips jpegsave documentation
• govips v2.19.0 JpegExportParams (vips/image.go)
• sharp jpeg() output options