Gary Lucas created IMAGING-312:
----------------------------------

             Summary: Alpha-channel setting not interpreted from ExtraSamples 
tag
                 Key: IMAGING-312
                 URL: https://issues.apache.org/jira/browse/IMAGING-312
             Project: Commons Imaging
          Issue Type: Bug
          Components: Format: TIFF
    Affects Versions: 1.0-alpha2
         Environment:  
            Reporter: Gary Lucas


Commons Imaging sometimes misinterprets TIFF files that have 4-byte RGB samples 
but do not define alpha.   In some cases, these images are treated as 
semi-transparent when they should be opaque.   Commons Imaging is not unique in 
this regard...  Windows Photo Viewer does the same thing.

The TIFF specification allows RGB images to be encoded with 4-bytes per pixel.  
It would be natural to assume (as Commons Imaging does) that the 4th byte is 
the alpha channel and that it would have values of 0xff in the case where 
pixels were opaque. However, the interpretation of the 4th byte depends on 
information in the TIFF "ExtraSamples" tag. 

It turns out that there are images in-the-wild that use 4 bytes, but populate 
the 4th byte with junk values. For example, there are a number of older aerial 
photographs from the US Geological Survey (USGS) that do this.  These images 
give an ExtraSamples tag with a value of zero.  But the TIFF specification 
calls for images to be treated as having alpha channels only if the 
ExtraSamples field carries a value of either 1 or 2.   When ExtraSamples has a 
value of 0, the 4th byte is to be ignored. 

There are many examples of this phenomenon on the USGS Earth Explorer website. 
One specific example: 

* High Resolution Orthoimagery
* Dataset: 201203_connecticut_state_lot1_ct_0x3000m_utm_cnir
* Entity: 2818289_18TYL425825
* File: 18tyl425825.tif 




*Proposed Fix*
I propose to do the following:
* Extend the TiffImageParser logic for detecting alpha to assume hasAlpha is 
true if and only if the ExtraSamples tag is supplied and contains values 1 or 
2. 
* Provide a hasAlpha accessor for the ImageBuilder class (is should really have 
one anyway)
* Enhance the DataReaderStrips and DataReaderTiles classes to check hasAlpha 
when processing RGB images that have 4 samples per pixel samples. 


*Concerns*
At this time, I am not sure what to do if an RGB TIFF image uses 4-samples per 
pixel but the ExtraSamples tag is not provided.  At this time, I have not seen 
an example of this, but my collection of sample TIFF files is rather narrow and 
I would not rule it out.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to