[
https://issues.apache.org/jira/browse/IMAGING-126?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13911567#comment-13911567
]
Gary Lucas commented on IMAGING-126:
------------------------------------
Thanks for the code sample.
It's been over a year since I've worked on Apache Commons Imaging and it's
taking me some time to get set up again. I've found the problem for TIFF files
and am looking into a solution. I imagine that the PNG files have the same
problem.
First off, I note that your image is 2550-by-3300 pixels in size or about 8.4
million pixels. At 3 bytes per pixel, the uncompressed image would be about 24
megabytes. Your output is about 1 megabyte, which is substantially smaller
than 24 M, but still bigger than it ought to be. The good news is that this
result indicates that some compression is happening. It's just not as
efficient as it should be.
In the class TiffImageWriterBase.java, there is a calculation on line 308 where
the writeImage method computes "rowsPerStrip". In effect, this is the number
of rows from the source image that the image writer will try to compress (each
"strip" is compressed independently). The code looks like this:
int rowsPerStrip = 64000 / (width * bitsPerSample * samplesPerPixel);
I think that 64000 is an arbitrarily selected number, but it is suspiciously
close to 64 K. I am not sure what the original author's intention was on this
one (he was not fond of adding comments to his code). I'll have to look at the
TIFF specification to see if there is a reason for this value.
Anyway, with your 2550-pixel-wide image, this calculation comes out to one row
per strip. Consequently, the LZW compressor compresses each row of pixels
independently. Of course, the LZW compressor is usually much better
compressing longer texts than it is at compressing short texts. Since the
image writer is taking only a single row of pixels at a time, the "texts" are
rather shorter than they could be. So the compressor does not achieve as good
ratios as it might.
I hacked the code to use a larger value than 64000 so that the rowsPerStrip
value was larger. This change reduced the size of the file quite a bit.
Unfortunately, now I have a block of black pixels added to the bottom of the
image... So there is something else about the way the image writer works that
I don't understand yet.
I'll let you know if I make more progress.
Gary
> TIFF and PNG images should not be bigger than the ones created by java ImageIO
> ------------------------------------------------------------------------------
>
> Key: IMAGING-126
> URL: https://issues.apache.org/jira/browse/IMAGING-126
> Project: Commons Imaging
> Issue Type: Improvement
> Components: Format: PNG, Format: TIFF
> Affects Versions: 1.0
> Environment: W7
> Reporter: Tilman Hausherr
> Priority: Minor
> Attachments: pdfbox-1870-devicen3-01.png,
> pdfbox-1870-devicen3-01.tif, pdfbox-1870-devicen3.pdf-1.png,
> pdfbox-1870-devicen3.pdf-1.tif
>
>
> I tried to use Apache Imaging for the PDFBOX project (PDFBOX-1734) because of
> problems with setting the tiff resolution in java imageio.
> While the code is pretty nice, I found that the generated images are
> sometimes much bigger in size than the ones generated by java imageio.
> Example:
> pdfbox-1870-devicen3-01.png 50 KB (imageio)
> pdfbox-1870-devicen3.pdf-1.png 70 KB (imaging)
> pdfbox-1870-devicen3-01.tif 401 KB (imageio)
> pdfbox-1870-devicen3.pdf-1.tif 1063 KB (imaging)
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
