jpegtran

Mirror an image horizontally or vertically

$ jpegtran [[-f|-flip]] [horizontal|vertical] [path/to/image.jpg] > [path/to/output.jpg]

$ jpegtran [[-ro|-rotate]] [90|180|270] [path/to/image.jpg] > [path/to/output.jpg]

$ jpegtran -transpose [path/to/image.jpg] > [path/to/output.jpg]

$ jpegtran -transverse [path/to/image.jpg] > [path/to/output.jpg]

$ jpegtran [[-g|-grayscale]] [path/to/image.jpg] > [path/to/output.jpg]

$ jpegtran -crop [W]x[H] -outfile [path/to/output.jpg] [path/to/image.jpg]

$ jpegtran -crop [W]x[H]+[X]+[Y] [path/to/image.jpg] > [path/to/output.jpg]

jpegtran [options] inputfile [outputfile]

-arithmetic
    Use arithmetic entropy coding instead of Huffman.

-copy none|comments|all|exif
    Copy specified APPn markers/metadata to output; default copies comments.

-crop WxH+x+y
    Losslessly crop to given widthxheight+xoffset+yoffset (iMCU-aligned).

-dct int|float
    Select DCT method: int (default, fast) or float (precise).

-drop none|any|all
    Drop non-APPn markers: none (default), any except JFIF/Adobe, or all.

-flip horizontal|vertical
    Mirror image horizontally or vertically.

-maxmemory N
    Limit memory usage to N megabytes.

-optimize
    Optimize entropy coding tables for smaller files.

-outfile name
    Write output to named file (default stdout).

-perfect
    Use perfect Huffman trees (slower, larger than -optimize).

-progressive
    Convert to baseline progressive JPEG.

-restart N
    Set restart interval to N MCU rows.

-rotate 90|180|270
    Rotate clockwise by 90, 180, or 270 degrees.

-scale 1/2|1/4|1/8
    Losslessly downscale by integer factor.

-transpose
    Transpose across UL-to-LR axis.

-transverse
    Transverse transpose across UR-to-LL axis.

-trim
    Trim off partial iMCU blocks at edges.

-verbose
    Enable verbose output.

jpegtran is a utility from the Independent JPEG Group's libjpeg library for performing lossless transformations on JPEG images. It allows operations like rotation, flipping, cropping, trimming, and scaling without re-encoding the image data, preserving original quality.

Key features include optimizing Huffman tables for smaller file sizes (-optimize), converting to progressive JPEG (-progressive), and handling metadata with -copy options. It supports downscaling by factors of 1/2, 1/4, or 1/8 (-scale), and can drop unnecessary markers to reduce size.

Unlike lossy tools, jpegtran rearranges the existing DCT coefficients, making it ideal for batch processing photos for web or storage. It reads from stdin or files and writes to stdout or specified output, with verbose mode for progress. Limitations apply to crop boundaries (must align with iMCU blocks), but -trim helps with ragged edges. Widely used in image workflows, it's faster and safer than re-encoding with converters like ImageMagick's mogrify.

Crops must align with iMCU grid (8-16 pixel blocks); use -trim for auto-adjust. No support for non-JPEG input. Optimized files may not be smaller if already optimal.

Optimize: jpegtran -optimize -outfile out.jpg in.jpg
Rotate: jpegtran -rotate 90 -copy all -outfile out.jpg in.jpg
Progressive: jpegtran -progressive -optimize input.jpg > output.jpg

Developed by Independent JPEG Group (IJG); first released in 1991 as part of libjpeg v5. Ongoing maintenance in libjpeg-turbo fork for speed improvements.

cjpeg(1), djpeg(1), rdjpgcom(1), wrjpgcom(1)