Decode barcodes on cropped specimen images

[quicklizard99]

Metadata fields need to be populated from data decoded from barcodes on cropped specimens images. Decoding should be robost and require as little human intervention as possible.

[quicklizard99]

Some good descriptions at Wikipedia and Inlite. Within the museum have least:

• Code 128 - 1d - herbarium sheets
• Data Matrix - 2d - microscope slides, iCollections

The ideal is we pass the cropped specimen image to a decoder, which returns the data encoded within (ideally also the types and locations) of all barcodes on the image.

OpenCV does not have barcode decoders. This has been slated for a while, e.g. for the 2014 GSoC with Claudia Rapuano as mentor. See also this discussion.

I have investigated the following open-source barcode decoders

zbar

• http://zbar.sourceforge.net/
• https://github.com/ZBar/ZBar/blob/master/python/README
• https://github.com/herbyme/zbar/blob/master/python/examples/scan_image.py
• Open-source library for Python
• Fast
• Can decode many different types of barcode but not handle DataMatrix (requested on at least two previous occasions)
  • http://sourceforge.net/p/zbar/discussion/1072195/thread/13240cf7/
  • http://sourceforge.net/p/zbar/discussion/664595/thread/25769670/
• No development since 2012
• Requires compilation

libdmtx

• http://www.libdmtx.org/
• http://sourceforge.net/p/libdmtx/libdmtx/ci/master/tree/
• Open-source cpp library
• Can be very slow - several 10s of seconds for a large image; it is possible to set a timeout
• Requires compilation
• No development since 2011 and not likely to be any soon
• Not a python library. We could create our own python wrapper.

zxing

• https://github.com/zxing/zxing
• Java library
• Used to have a cpp version but it appears that this is no longer maintained
• A QT port has been made

OpenCV + decoder library

None of these three open-source decoders is capable of reliably locating and decoding barcodes within the example specimen images that I have.

I coded up this strategy:

• Use OpenCV to identify areas of the image that might contain barcodes
• Pass each candidate area through one or more of the open-source decoders

This method works very well for some images by not reliably across the example cropped specimen images that I have.

Next step

Look at commercial decoders.

[stefanv]

Could you place a selection of images online for which this should work? It would help a lot in coding up some test solutions.

[quicklizard99]

Thanks @stefanv. I will share a few hundred example images, along with some code and results, although probably not until the start of next week.

[quicklizard99]

The open-source decoders are very slow and/or fail to find barcodes - commercial libraries give much better results. Inselect is currently hard-coded to use the Inlite ClearImage library. The library only runs on Windows - if you are using a Mac or Linux then we currently have no good solution. ClearImage supports a wide range of barcode types but Inselect is hard-coded to detect only Data Matrix. See also #130.

[quicklizard99]

NaturalHistoryMuseum/gouda@f8f2294 adds supports for libdmtx - tested on Mac OS X only.

[quicklizard99]

libdmtx via pydmtx also tested on Linux and 32-bit Windows. 64-bit windows build of libdmtx fails with core dump.

[quicklizard99]

libdmtx via pydmtx working on Mac OS X, Linux and 32-bit and 64-bit Windows.

[quicklizard99]

v0.1.8 (f8f0667) introduces support for the open-source libdmtx decoder on all platforms. See
NaturalHistoryMuseum/libdmtx and NaturalHistoryMuseum/dmtx-wrappers.

[quicklizard99]

Adding support for the zbar open-source library would allow 1D barcodes and QR codes to be decoded. The last release (0.10) was on 2009-10-27. There has been no development since and not likely to be any more. zbar is available for all three major OSes but, frustratingly, there is no 64-bit build for Windows - a show stopper.

We would need to

• add support for 64-bit Windows
• add a user interface config box that allows the user to choose between
  • libdmtx
  • Inlite
  • zbar

[quicklizard99]

User interface for config will be handled by #156

[quicklizard99]

Support for QR requested by AMNH. Will require that Inselect supports zbar and/or exposes the relevant Inlite functionality.

[quicklizard99]

Inselect now supports Inlite, zbar and libdmtx. The only gap is that zbar is not available on 64-bit Windows, for reasons given on 2015-04-20. Edit, Configure Read barcodes shows a box allowing the user to choose between the three decoders; options are disabled where a decoder is unavailable:

[quicklizard99]

Our fork of ZBarWin64 adds support for zbar on 64-bit Windows.