gdal - gdal_translate- why jpg compressed tif is 2 times greater than jpg file?

Using the following two commands yields completely different file sizes:

gdal_translate test.tif -of GTiff -co "COMPRESS=JPEG" -co "JPEG_QUALITY=100" output.tif

and

gdal_translate test.tif -of JPEG -co "QUALITY=100" -co "WORLDFILE=YES" output.jpg.

(I Wrote the commands from my memory, not sure that the syntax is correct)

The jpg file (second command) is 2 times smaller than the compressed tif file. Why?

Answer

By default an RGB image will be written to an RGB color model JPEG image, but this is not actually the most efficient way of writing to JPEG. It is better to convert to the YCbCr color space, and encode that. This is in fact the typical form of standalone JPEGs and what GDAL will produce when writing to a free standing JPEG file. Compressing a 4K x 2.6K image I see:

gdal_translate rgb.tif out.tif -co COMPRESS=JPEG 
$ ls -l out.tif
-rw-rw-r-- 1 warmerdam warmerdam 7416934 Nov  4 10:09 out.tif 
gdal_translate rgb.tif out_ycbcr.tif -co COMPRESS=JPEG -co PHOTOMETRIC=YCBCR
$ ls -l out_ycbcr.tif
-rw-rw-r-- 1 warmerdam warmerdam 2251264 Nov  4 10:09 out_ycbcr.tif

With YCbCr encoding the size of the output is pretty much the same as the JFIF JPEG image.