[mapserver-users] Image compression/performance

[Stephen Woodbridge]

This is good information, thanks. I don't think jpeg in tiff was an 
option when I did this some years ago, but if is clearly a good trick if 
you can deal with the lossiness of jpeg compression.

-SteveW

On 3/22/2012 4:03 AM, Rahkonen Jukka wrote:
> Stephen Woodbridge wrote:
>
>> On 3/21/2012 11:18 PM, Ian Walberg wrote:
>>> Folks,
>>>
>>> We are using tif format images and getting good map rendering
>>> performance.
>>>
>>> However the image file size could do with reducing a little.
>>>
>>> Anyone got experience of what compression options we have
>> that have the
>>> least impact on performance?
>>
>> My experience of working with USGS DOQQ satellite imagery in GeoTiff
>> files, was that the amount of compression really did not make a
>> significant reduction in size. We used internally tiled tiff
>> files with
>> overviews and had over 15TB of imagery online. There are
>> other formats
>> that provide higher compression rates like MrSID and others, but
>> depending on the format the behavior characteristics vary
>> greatly and I
>> do not have any recent stats or comparisons of size versus
>> performance
>> versus format.
>
>
> Hi,
>
> I took numbers from one of our orthoimage set
>
> Uncompressed tiff files from the contractor
>   667 File(s) 289 260 585 944 bytes
>
> What we have on server disk after running
> gdal_translate -co COMPRESS=JPEG -co PHOTOMETRIC=YCBCR
>   667 File(s) 21 411 672 702 bytes
>
> Reduction in size 93%.  Aerial images are used for on-screen interpretation and difference in quality is not visible with bare eyes. The rate of the JPEG compression can be adjusted but I have been happy with GDAL defaults. Overviews seem to add about 40-50% to JPEG compressed images. Those I have created probably as
> gdaladdo -ro --config COMPRESS_OVERVIEW JPEG --config PHOTOMETRIC_OVERVIEW YCBCR
>           --config INTERLEAVE_OVERVIEW PIXEL tiff.tif 2 4 8 16 32 64
>
> We have also JPEG2000 versions of the images but GDAL does not handle them as fast as JPEG-in-TIFF files. A few years ago the speed with JPEG2000 (with EWCJP2 and KAKJP2 drivers) was about the half of what we got with tiffs. Those numbers are nothing to rely on today because the software versions have changed so many times.  Because of the unfriendly licenses of the good JPEG2000 libraries and because jpeg-in-tiff works so well I have not bothered to repeat the tests myself lately. Generally, for this kind of questions the best answer is achieved by making a well controlled test in your own environment. You can then publish your results and tell how you did the test so that others can check if something has been were poorly configured. For example it is not at all the same how JPEG2000 images have been compressed and JPEG compressed tiffs without tiles will for sure be a fiasco.
>
> -Jukka Rahkonen-
>   _______________________________________________
> mapserver-users mailing list
> mapserver-users at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/mapserver-users