Examples

Basic Image I/O

How do I…?

… read images?

Jp2k implements slicing via the __getitem__() method and hooks it into the multiple resolution property of JPEG 2000 imagery. This allows you to retrieve multiresolution imagery via array-style slicing, i.e. strides. For example, here’s how to retrieve a full resolution, first lower-resolution image, and second lower-resolution image. It’s always powers of two.

>>> import glymur
>>> j2kfile = glymur.data.goodstuff() # just a path to a raw JPEG 2000 codestream
>>> j2k = glymur.Jp2k(j2kfile)
>>> fullres = j2k[:]
>>> fullres.shape
(800, 480, 3)
>>> thumbnail = j2k[::2, ::2]
>>> thumbnail.shape
(400, 240, 3)
>>> thumbnail2 = j2k[::4, ::4]
>>> thumbnail.shape
(200, 120, 3)

… read really large images

JPEG 2000 images can be much larger than what can fit into your computer’s memory. While you can use strides that align with the JPEG 2000 decomposition levels to retrieve lower resolution images, retrieving the lowest resolution image would seem to require that you know just how many decomposition levels are available. While you can get that information from the COD segment in the codestream, glymur provides you with a shortcut. Normally the stride must be a power of 2, but you can provide -1 instead to get the smallest thumbnail.:

>>> import glymur
>>> j2kfile = glymur.data.goodstuff() # just a path to a JPEG 2000 file
>>> j2k = glymur.Jp2k(j2kfile)
>>> j2k.shape
(800, 480, 3)
>>> j2k.codestream.segment[2].num_res
5
>>> j2k[::32, ::32].shape
(25, 15, 3)
>>> thumbnail = j2k[::-1, ::-1] # last thumbnail was the 5th, 2 ** 5 = 32
>>> thumbnail.shape
(25, 15, 3)

… read an image layer?

JPEG2000 has layers which allow you to specify images with different levels of quality. Different layers may be specified by utilizing the layer property. The default layer value is 0, which specifies the first layer.

>>> import glymur
>>> file = glymur.data.jpxfile() # just a path to a JPEG2000 file
>>> jpx = glymur.Jp2k(file)
>>> d0 = jpx[:] # first layer
>>> jpx.layer = 3
>>> d3 = jpx[:] # third layer

… write images?

The easiest way is just to assign the entire image, similar to what might be done with NumPy.

>>> import glymur, skimage.data
>>> jp2 = glymur.Jp2k('astronaut.jp2')
>>> jp2[:] = skimage.data.astronaut()

Advanced Image I/O

How do I…?

… make use of OpenJPEG’s thread support to read images?

If you have glymur 0.8.13 or higher and OpenJPEG 2.2.0 or higher, you can make use of OpenJPEG’s thread support to speed-up read operations. If you have really big images and a large number of cores at your disposal, you really should look into this.

>>> import glymur
>>> import time
>>> jp2file = glymur.data.nemo()
>>> jp2 = glymur.Jp2k(jp2file)
>>> t0 = time.time(); data = jp2[:]; t1 = time.time()
>>> t1 - t0
0.9024193286895752
>>> glymur.set_option('lib.num_threads', 2)
>>> t0 = time.time(); data = jp2[:]; t1 = time.time()
>>> t1 - t0
0.4060473537445068

… efficiently read just one band of a big image?

For really large images, before v0.9.4 you had to read in all bands of an image, even if you were only interested in just one of those bands. With v0.9.4 or higher, you can make use of the decoded_components() property, which will inform the openjpeg library to just decode the specified component(s), which can significantly speed up read operations on large images. Be aware, however, that the openjpeg library will not employ the MCT when decoding these components.

You can set the property to None to restore the behavior of decoding all bands.

>>> import glymur
>>> jp2file = glymur.data.nemo()
>>> jp2 = glymur.Jp2k(jp2file)
>>> data = jp2[:]
>>> data.shape
(1456, 2592, 3)
>>> jp2.decoded_components = 1
>>> data = jp2[:]
>>> data.shape
(1456, 2592)
>>> jp2.decoded_components = [0, 2]
>>> data = jp2[:]
>>> data.shape
(1456, 2592, 2)
>>> jp2.decoded_components = None
>>> data = jp2[:]
>>> data.shape
(1456, 2592, 3)

… write images using multithreaded encoding?

If you have glymur 0.9.3 or higher and OpenJPEG 2.4.0 or higher, you can make use of OpenJPEG’s thread support to speed-up read operations. With a puny 2015 macbook, just two cores, and a 5824x10368x3 image, we get:

>>> import time, numpy as np, glymur
>>> data = glymur.Jp2k(glymur.data.nemo())[:]
>>> data = np.tile(data, (4, 4, 1))
>>> t0 = time.time()
>>> glymur.Jp2k('1thread.jp2', data)
>>> t1 = time.time()
>>> print(f'1 thread:  {(t1 - t0):.3} seconds')
12.0 seconds
>>> t0 = time.time()
>>> glymur.set_option('lib.num_threads', 2)
>>> glymur.Jp2k('2threads.jp2', data)
>>> t1 = time.time()
>>> print(f'2 threads:  {(t1 - t0):.3} seconds')
7.24 seconds

… write images that cannot fit into memory?

If you have glymur 0.9.4 or higher, you can write out an image tile-by-tile. In this example, we take a 512x512x3 image and tile it into a 20x20 grid, resulting in a 10240x10240x3 image.

>>> import skimage.data
>>> from glymur import Jp2k
>>> img = skimage.data.astronaut()
>>> print(img.shape)
(512, 512, 3)
>>> shape = img.shape[0] * 20, img.shape[1] * 20, 3
>>> tilesize = (img.shape[0], img.shape[1])
>>> j = Jp2k('400astronauts.jp2', shape=shape, tilesize=tilesize, verbose=True)
>>> for tw in j.get_tilewriters():
        tw[:] = img
>>> j = Jp2k('400astronauts.jp2')
>>> print(j.shape)
(10240, 10240, 3)

Note that the tiles are written out left-to-right, tile-row-by-tile-row. You must have image data ready to feed each tile writer, you cannot skip a tile.

… force the generation of PLT markers?

With glymur 0.9.5 or higher, you can instruct the encoder to generate PLT markers by using the plt keyword.

>>> import glymur, skimage.data
>>> jp2 = glymur.Jp2k('astronaut.jp2', plt=True)
>>> jp2[:] = skimage.data.astronaut()
>>> c = jp2.get_codestream(header_only=False)
>>> print(c.segment[6])
PLT marker segment @ (222, 45)
    Index:  0
    Iplt:  [271, 201, 208, 749, 551, 548, 2569, 1852, 1814, 8300, 6370, 6061, 26987, 23437, 21431, 88511, 86763, 77253]

… write images with different compression ratios for different layers?

Different compression factors may be specified with the cratios parameter

>>> import skimage.data, glymur
>>> data = skimage.data.camera()
>>> # quality layer 1: compress 20x
>>> # quality layer 2: compress 10x
>>> # quality layer 3: compress lossless
>>> jp2 = glymur.Jp2k('myfile.jp2', data=data, cratios=[20, 10, 1])
>>> # read the lossless layer
>>> jp2.layer = 2
>>> data = jp2[:]

… write images with different PSNR (or “quality”) for different layers?

Different PSNR values may be specified with the psnr parameter. Please read https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio for a basic understanding of PSNR.

Values must be increasing, but the last value may be 0 to indicate the layer is lossless. However, the OpenJPEG library will reorder the layers to make the first layer lossless, not the last.

>>> import skimage.data, skimage.metrics, glymur
>>> truth = skimage.data.camera()
>>> jp2 = glymur.Jp2k('myfile.jp2', data=truth, psnr=[30, 40, 50, 0])
>>> psnr = []
>>> for layer in range(4):
...     jp2.layer = layer
...     psnr.append(skimage.metrics.peak_signal_noise_ratio(truth, jp2[:]))
>>> print(psnr)
[inf, 29.028560403833303, 39.206919416670402, 47.593129828702246]

… convert TIFF images to JPEG 2000?

Many TIFFs can be converted to tiled JPEG 2000 files using glymur. A command line utility tiff2jp2 is provided for this task.

In most cases, you should provide your own choice of a JPEG 2000 tile size. Not providing a tile size will cause glymur to try to covert the TIFF into a single-tile JPEG 2000 file. If your TIFF is large, you may not have enough memory to write such a single-tile file.

$ wget http://photojournal.jpl.nasa.gov/tiff/PIA17145.tif
$ tiff2jp2 --tilesize 256 256 PIA17145.tif PIA17145.jp2

If your TIFF is really big but has an unfortunate choice for the RowsPerStrip tag (like the seemingly ubiquitous value of 3, which was reasonable only in prehistoric times) … well that’s going to be very inefficient no matter how you tile the JPEG 2000 file.

The TIFF metadata is stored in a UUID box appended to the end of the JPEG 2000 file.

… create an image with an alpha layer?

OpenJPEG can create JP2 files with more than 3 components (use version 2.1.0+ for this), but by default, any extra components are not described as such. In order to do so, we need to re-wrap such an image in a set of boxes that includes a channel definition box. The following example creates an ellipical mask.

>>> import numpy as np
>>> import glymur
>>> from glymur import Jp2k
>>> rgb = Jp2k(glymur.data.goodstuff())[:]
>>> ny, nx = rgb.shape[:2]
>>> Y, X = np.ogrid[:ny, :nx]
>>> mask = nx ** 2 * (Y - ny / 2) ** 2 + ny ** 2 * (X - nx / 2) ** 2 > (nx * ny / 2)**2
>>> alpha = 255 * np.ones((ny, nx, 1), dtype=np.uint8)
>>> alpha[mask] = 0
>>> rgba = np.concatenate((rgb, alpha), axis=2)
>>> jp2 = Jp2k('myfile.jp2', data=rgba)

Next we need to specify what types of channels we have. The first three channels are color channels, but we identify the fourth as an alpha channel:

>>> from glymur.core import COLOR, OPACITY
>>> ctype = [COLOR, COLOR, COLOR, OPACITY]

And finally we have to specify just exactly how each channel is to be interpreted. The color channels are straightforward, they correspond to R-G-B, but the alpha (or opacity) channel in this case is to be applied against the entire image (it is possible to apply an alpha channel to a single color channel, but we aren’t doing that).

>>> from glymur.core import RED, GREEN, BLUE, WHOLE_IMAGE
>>> asoc = [RED, GREEN, BLUE, WHOLE_IMAGE]
>>> cdef = glymur.jp2box.ChannelDefinitionBox(ctype, asoc)
>>> print(cdef)
Channel Definition Box (cdef) @ (0, 0)
    Channel 0 (color) ==> (1)
    Channel 1 (color) ==> (2)
    Channel 2 (color) ==> (3)
    Channel 3 (opacity) ==> (whole image)

It’s easiest to take the existing jp2 jacket and just add the channel definition box in the appropriate spot. The channel definition box must go into the jp2 header box, and then we can rewrap the image.

>>> boxes = jp2.box  # The box attribute is the list of JP2 boxes
>>> boxes[2].box.append(cdef)
>>> jp2_rgba = jp2.wrap("goodstuff_rgba.jp2", boxes=boxes)

Here’s how the Preview application on the mac shows the RGBA image.

_images/goodstuff_alpha.png

Basic Metadata

How do I…?

… display metadata?

There are two ways. From the command line, the console script jp2dump is available.

$ jp2dump /path/to/glymur/installation/data/nemo.jp2

From within Python, the same result is obtained simply by printing the Jp2k object, i.e.

>>> import glymur
>>> jp2file = glymur.data.nemo() # just a path to a JP2 file
>>> jp2 = glymur.Jp2k(jp2file)
>>> print(jp2)
File:  nemo.jp2
JPEG 2000 Signature Box (jP  ) @ (0, 12)
    Signature:  0d0a870a
File Type Box (ftyp) @ (12, 20)
    Brand:  jp2
    Compatibility:  ['jp2 ']
JP2 Header Box (jp2h) @ (32, 45)
    Image Header Box (ihdr) @ (40, 22)
        Size:  [1456 2592 3]
        Bitdepth:  8
        Signed:  False
        Compression:  wavelet
        Colorspace Unknown:  False
    Colour Specification Box (colr) @ (62, 15)
        Method:  enumerated colorspace
        Precedence:  0
        Colorspace:  sRGB
UUID Box (uuid) @ (77, 3146)
    UUID:  be7acfcb-97a9-42e8-9c71-999491e3afac (XMP)
    UUID Data:
    <ns0:xmpmeta xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:ns0="adobe:ns:meta/" xmlns:ns2="http://ns.adobe.com/xap/1.0/" xmlns:ns3="http://ns.adobe.com/tiff/1.0/" xmlns:ns4="http://ns.adobe.com/exif/1.0/" xmlns:ns5="http://ns.adobe.com/photoshop/1.0/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" ns0:xmptk="Exempi + XMP Core 5.1.2">
      <rdf:RDF>
        <rdf:Description rdf:about="">
          <ns2:CreatorTool>Google</ns2:CreatorTool>
          <ns2:CreateDate>2013-02-09T14:47:53</ns2:CreateDate>
        </rdf:Description>
        <rdf:Description rdf:about="">
          <ns3:YCbCrPositioning>1</ns3:YCbCrPositioning>
          <ns3:XResolution>72/1</ns3:XResolution>
          <ns3:YResolution>72/1</ns3:YResolution>
          <ns3:ResolutionUnit>2</ns3:ResolutionUnit>
          <ns3:Make>HTC</ns3:Make>
          <ns3:Model>HTC Glacier</ns3:Model>
          <ns3:ImageWidth>2592</ns3:ImageWidth>
          <ns3:ImageLength>1456</ns3:ImageLength>
          <ns3:BitsPerSample>
            <rdf:Seq>
              <rdf:li>8</rdf:li>
              <rdf:li>8</rdf:li>
              <rdf:li>8</rdf:li>
            </rdf:Seq>
          </ns3:BitsPerSample>
          <ns3:PhotometricInterpretation>2</ns3:PhotometricInterpretation>
          <ns3:SamplesPerPixel>3</ns3:SamplesPerPixel>
          <ns3:WhitePoint>
            <rdf:Seq>
              <rdf:li>1343036288/4294967295</rdf:li>
              <rdf:li>1413044224/4294967295</rdf:li>
            </rdf:Seq>
          </ns3:WhitePoint>
          <ns3:PrimaryChromaticities>
            <rdf:Seq>
              <rdf:li>2748779008/4294967295</rdf:li>
              <rdf:li>1417339264/4294967295</rdf:li>
              <rdf:li>1288490240/4294967295</rdf:li>
              <rdf:li>2576980480/4294967295</rdf:li>
              <rdf:li>644245120/4294967295</rdf:li>
              <rdf:li>257698032/4294967295</rdf:li>
            </rdf:Seq>
          </ns3:PrimaryChromaticities>
        </rdf:Description>
        <rdf:Description rdf:about="">
          <ns4:ColorSpace>1</ns4:ColorSpace>
          <ns4:PixelXDimension>2528</ns4:PixelXDimension>
          <ns4:PixelYDimension>1424</ns4:PixelYDimension>
          <ns4:FocalLength>353/100</ns4:FocalLength>
          <ns4:GPSAltitudeRef>0</ns4:GPSAltitudeRef>
          <ns4:GPSAltitude>0/1</ns4:GPSAltitude>
          <ns4:GPSMapDatum>WGS-84</ns4:GPSMapDatum>
          <ns4:DateTimeOriginal>2013-02-09T14:47:53</ns4:DateTimeOriginal>
          <ns4:ISOSpeedRatings>
            <rdf:Seq>
              <rdf:li>76</rdf:li>
            </rdf:Seq>
          </ns4:ISOSpeedRatings>
          <ns4:ExifVersion>0220</ns4:ExifVersion>
          <ns4:FlashpixVersion>0100</ns4:FlashpixVersion>
          <ns4:ComponentsConfiguration>
            <rdf:Seq>
              <rdf:li>1</rdf:li>
              <rdf:li>2</rdf:li>
              <rdf:li>3</rdf:li>
              <rdf:li>0</rdf:li>
            </rdf:Seq>
          </ns4:ComponentsConfiguration>
          <ns4:GPSLatitude>42,20.56N</ns4:GPSLatitude>
          <ns4:GPSLongitude>71,5.29W</ns4:GPSLongitude>
          <ns4:GPSTimeStamp>2013-02-09T19:47:53Z</ns4:GPSTimeStamp>
          <ns4:GPSProcessingMethod>NETWORK</ns4:GPSProcessingMethod>
        </rdf:Description>
        <rdf:Description rdf:about="">
          <ns5:DateCreated>2013-02-09T14:47:53</ns5:DateCreated>
        </rdf:Description>
        <rdf:Description rdf:about="">
          <dc:Creator>
            <rdf:Seq>
              <rdf:li>Glymur</rdf:li>
              <rdf:li>Python XMP Toolkit</rdf:li>
            </rdf:Seq>
          </dc:Creator>
        </rdf:Description>
      </rdf:RDF>
    </ns0:xmpmeta>
Contiguous Codestream Box (jp2c) @ (3223, 1132296)
    Main header:
        SOC marker segment @ (3231, 0)
        SIZ marker segment @ (3233, 47)
            Profile:  2
            Reference Grid Height, Width:  (1456 x 2592)
            Vertical, Horizontal Reference Grid Offset:  (0 x 0)
            Reference Tile Height, Width:  (1456 x 2592)
            Vertical, Horizontal Reference Tile Offset:  (0 x 0)
            Bitdepth:  (8, 8, 8)
            Signed:  (False, False, False)
            Vertical, Horizontal Subsampling:  ((1, 1), (1, 1), (1, 1))
        COD marker segment @ (3282, 12)
            Coding style:
                Entropy coder, without partitions
                SOP marker segments:  False
                EPH marker segments:  False
            Coding style parameters:
                Progression order:  LRCP
                Number of layers:  2
                Multiple component transformation usage:  reversible
                Number of resolutions:  2
                Code block height, width:  (64 x 64)
                Wavelet transform:  5-3 reversible
                Precinct size:  default, 2^15 x 2^15
                Code block context:
                    Selective arithmetic coding bypass:  False
                    Reset context probabilities on coding pass boundaries:  False
                    Termination on each coding pass:  False
                    Vertically stripe causal context:  False
                    Predictable termination:  False
                    Segmentation symbols:  False
        QCD marker segment @ (3296, 7)
            Quantization style:  no quantization, 2 guard bits
            Step size:  [(0, 8), (0, 9), (0, 9), (0, 10)]
        CME marker segment @ (3305, 37)
            "Created by OpenJPEG version 2.0.0"

… display less metadata?

The amount of metadata in a JPEG 2000 file can be overwhelming, mostly due to the codestream and XML and UUID boxes. You can suppress the codestream and XML details by making use of the set_option() function:

>>> glymur.set_option('print.codestream', False)
>>> glymur.set_option('print.xml', False)
>>> print(jp2)
File:  nemo.jp2
JPEG 2000 Signature Box (jP  ) @ (0, 12)
    Signature:  0d0a870a
File Type Box (ftyp) @ (12, 20)
    Brand:  jp2
    Compatibility:  ['jp2 ']
JP2 Header Box (jp2h) @ (32, 45)
    Image Header Box (ihdr) @ (40, 22)
        Size:  [1456 2592 3]
        Bitdepth:  8
        Signed:  False
        Compression:  wavelet
        Colorspace Unknown:  False
    Colour Specification Box (colr) @ (62, 15)
        Method:  enumerated colorspace
        Precedence:  0
        Colorspace:  sRGB
UUID Box (uuid) @ (77, 3146)
    UUID:  be7acfcb-97a9-42e8-9c71-999491e3afac (XMP)
Contiguous Codestream Box (jp2c) @ (3223, 1132296)

… display the codestream in all its gory glory?

The codestream details are limited to the codestream header because by default that’s all the codestream metadata that is retrieved. It is, howver, possible to print the full codestream.:

>>> glymur.set_option('print.codestream', True)
>>> c = j.get_codestream(header_only=False)
>>> print(c)
Codestream:
SOC marker segment @ (3231, 0)
SIZ marker segment @ (3233, 47)
    Profile:  no profile
    Reference Grid Height, Width:  (1456 x 2592)
    Vertical, Horizontal Reference Grid Offset:  (0 x 0)
    Reference Tile Height, Width:  (1456 x 2592)
    Vertical, Horizontal Reference Tile Offset:  (0 x 0)
    Bitdepth:  (8, 8, 8)
    Signed:  (False, False, False)
    Vertical, Horizontal Subsampling:  ((1, 1), (1, 1), (1, 1))
COD marker segment @ (3282, 12)
    Coding style:
        Entropy coder, without partitions
        SOP marker segments:  False
        EPH marker segments:  False
    Coding style parameters:
        Progression order:  LRCP
        Number of layers:  2
        Multiple component transformation usage:  reversible
        Number of decomposition levels:  1
        Code block height, width:  (64 x 64)
        Wavelet transform:  5-3 reversible
        Precinct size:  (32768, 32768)
        Code block context:
            Selective arithmetic coding bypass:  False
            Reset context probabilities on coding pass boundaries:  False
            Termination on each coding pass:  False
            Vertically stripe causal context:  False
            Predictable termination:  False
            Segmentation symbols:  False
QCD marker segment @ (3296, 7)
    Quantization style:  no quantization, 2 guard bits
    Step size:  [(0, 8), (0, 9), (0, 9), (0, 10)]
CME marker segment @ (3305, 37)
    "Created by OpenJPEG version 2.0.0"
SOT marker segment @ (3344, 10)
    Tile part index:  0
    Tile part length:  1132173
    Tile part instance:  0
    Number of tile parts:  1
COC marker segment @ (3356, 9)
    Associated component:  1
    Coding style for this component:  Entropy coder, PARTITION = 0
    Coding style parameters:
        Number of decomposition levels:  1
        Code block height, width:  (64 x 64)
        Wavelet transform:  5-3 reversible
        Precinct size:  (32768, 32768)
        Code block context:
            Selective arithmetic coding bypass:  False
            Reset context probabilities on coding pass boundaries:  False
            Termination on each coding pass:  False
            Vertically stripe causal context:  False
            Predictable termination:  False
            Segmentation symbols:  False
QCC marker segment @ (3367, 8)
    Associated Component:  1
    Quantization style:  no quantization, 2 guard bits
    Step size:  [(0, 8), (0, 9), (0, 9), (0, 10)]
COC marker segment @ (3377, 9)
    Associated component:  2
    Coding style for this component:  Entropy coder, PARTITION = 0
    Coding style parameters:
        Number of decomposition levels:  1
        Code block height, width:  (64 x 64)
        Wavelet transform:  5-3 reversible
        Precinct size:  (32768, 32768)
        Code block context:
            Selective arithmetic coding bypass:  False
            Reset context probabilities on coding pass boundaries:  False
            Termination on each coding pass:  False
            Vertically stripe causal context:  False
            Predictable termination:  False
            Segmentation symbols:  False
QCC marker segment @ (3388, 8)
    Associated Component:  2
    Quantization style:  no quantization, 2 guard bits
    Step size:  [(0, 8), (0, 9), (0, 9), (0, 10)]
SOD marker segment @ (3398, 0)
EOC marker segment @ (1135517, 0)

Advanced Metadata

How do I…?

… add XML metadata?

You can append any number of XML boxes to a JP2 file (not to a raw codestream). Consider the following XML file data.xml :

<?xml version="1.0"?>
<info>
    <locality>
        <city>Boston</city>
        <snowfall>24.9 inches</snowfall>
    </locality>
    <locality>
        <city>Portland</city>
        <snowfall>31.9 inches</snowfall>
    </locality>
    <locality>
        <city>New York City</city>
        <snowfall>11.4 inches</snowfall>
    </locality>
</info>

The append() method can add an XML box as shown below:

>>> import shutil
>>> import glymur
>>> shutil.copyfile(glymur.data.nemo(), 'myfile.jp2')
>>> jp2 = glymur.Jp2k('myfile.jp2')
>>> xmlbox = glymur.jp2box.XMLBox(filename='data.xml')
>>> jp2.append(xmlbox)
>>> print(jp2)

… create display and/or capture resolution boxes?

Capture and display resolution boxes are part of the JPEG 2000 standard. You may create such metadata boxes via keyword arguments.:

>>> import numpy as np, glymur
>>> vresc, hresc = 0.1, 0.2
>>> vresd, hresd = 0.3, 0.4
>>> j = glymur.Jp2k(
        'my.jp2', data=np.zeros([256, 256, 3], dtype=np.uint8),
        capture_resolution=[vresc, hresc],
        display_resolution=[vresd, hresd],
    )
>>> glymur.set_printoptions(short=True)
>>> print(j)
File:  my.jp2
JPEG 2000 Signature Box (jP  ) @ (0, 12)
File Type Box (ftyp) @ (12, 20)
JP2 Header Box (jp2h) @ (32, 89)
    Image Header Box (ihdr) @ (40, 22)
    Colour Specification Box (colr) @ (62, 15)
    Resolution Box (res ) @ (77, 44)
        Capture Resolution Box (resc) @ (85, 18)
        Display Resolution Box (resd) @ (103, 18)
Contiguous Codestream Box (jp2c) @ (121, 174)

… reinterpret a codestream (say what)?

An existing raw codestream (or JP2 file) can be re-wrapped in a user-defined set of JP2 boxes. The point to doing this might be to provide a different interpretation of an image. For example, a raw codestream has no concept of a color model, whereas a JP2 file with a 3-channel codestream will by default consider that to be an RGB image.

To get just a minimal JP2 jacket on the codestream provided by goodstuff.j2k (a file consisting of a raw codestream), you can use the wrap() method with no box argument:

>>> import glymur
>>> glymur.set_option('print.codestream', False)
>>> jp2file = glymur.data.goodstuff()
>>> j2k = glymur.Jp2k(jp2file)
>>> jp2 = j2k.wrap("newfile.jp2")
>>> print(jp2)
File:  newfile.jp2
JPEG 2000 Signature Box (jP  ) @ (0, 12)
    Signature:  0d0a870a
File Type Box (ftyp) @ (12, 20)
    Brand:  jp2
    Compatibility:  ['jp2 ']
JP2 Header Box (jp2h) @ (32, 45)
    Image Header Box (ihdr) @ (40, 22)
        Size:  [800 480 3]
        Bitdepth:  8
        Signed:  False
        Compression:  wavelet
        Colorspace Unknown:  False
    Colour Specification Box (colr) @ (62, 15)
        Method:  enumerated colorspace
        Precedence:  0
        Colorspace:  sRGB
Contiguous Codestream Box (jp2c) @ (77, 115228)

The raw codestream was wrapped in a JP2 jacket with four boxes in the outer layer (the signature, file type, JP2 header, and contiguous codestream), with two additional boxes (image header and color specification) contained in the JP2 header superbox.

XML boxes are not in the minimal set of box requirements for the JP2 format, so in order to add an XML box into the mix before the codestream box, we’ll need to re-specify all of the boxes. If you already have a JP2 jacket in place, you can just reuse that, though. Take the following example content in an XML file favorites.xml :

<?xml version="1.0"?>
<favorite_things>
    <category>Light Ale</category>
</favorite_things>

In order to add the XML after the JP2 header box, but before the codestream box, the following will work.

>>> boxes = jp2.box  # The box attribute is the list of JP2 boxes
>>> xmlbox = glymur.jp2box.XMLBox(filename='favorites.xml')
>>> boxes.insert(3, xmlbox)
>>> jp2_xml = jp2.wrap("newfile_with_xml.jp2", boxes=boxes)
>>> print(jp2_xml)
File:  newfile_with_xml.jp2
JPEG 2000 Signature Box (jP  ) @ (0, 12)
    Signature:  0d0a870a
File Type Box (ftyp) @ (12, 20)
    Brand:  jp2
    Compatibility:  ['jp2 ']
JP2 Header Box (jp2h) @ (32, 45)
    Image Header Box (ihdr) @ (40, 22)
        Size:  [800 480 3]
        Bitdepth:  8
        Signed:  False
        Compression:  wavelet
        Colorspace Unknown:  False
    Colour Specification Box (colr) @ (62, 15)
        Method:  enumerated colorspace
        Precedence:  0
        Colorspace:  sRGB
XML Box (xml ) @ (77, 76)
    <favorite_things>
      <category>Light Ale</category>
    </favorite_things>
Contiguous Codestream Box (jp2c) @ (153, 115236)

As to the question of which method you should use, append() or wrap(), to add metadata, you should keep in mind that wrap() produces a new JP2 file, while append() modifies an existing file and is currently limited to XML and UUID boxes.

… work with ICC profiles?

A detailed answer is beyond my capabilities. What I can tell you is how to gain access to ICC profiles that JPEG 2000 images may or may not provide for you. If there is an ICC profile, it will be provided in a ColourSpecification box, but only if the colorspace attribute is None. Here is an example of how you can access an ICC profile in an example JPX file. Basically what is done is that the raw bytes corresponding to the ICC profile are wrapped in a BytesIO object, which is fed to the most-excellent Pillow package.

>>> from glymur import Jp2k
>>> from PIL import ImageCms
>>> from io import BytesIO
>>> # This next step produces a harmless warning that has nothing to do with ICC profiles.
>>> j = Jp2k('text_GBR.jp2')
>>> # The 2nd sub box of the 4th box is a ColourSpecification box.
>>> print(j.box[3].box[1].colorspace)
None
>>> b = BytesIO(j.box[3].box[1].icc_profile)
>>> icc = ImageCms.ImageCmsProfile(b)

To go any further with this, you will want to consult the Pillow documentation.

… work with XMP UUIDs?

Wikipedia states that “The Extensible Metadata Platform (XMP) is an ISO standard, originally created by Adobe Systems Inc., for the creation, processing and interchange of standardized and custom metadata for all kinds of resources.”

The example JP2 file shipped with glymur has an XMP UUID.

>>> import glymur
>>> j = glymur.Jp2k(glymur.data.nemo())
>>> print(j.box[3]) # formatting added to the XML below
<ns0:xmpmeta xmlns:dc="http://purl.org/dc/elements/1.1/"
             xmlns:ns0="adobe:ns:meta/"
             xmlns:ns2="http://ns.adobe.com/xap/1.0/"
             xmlns:ns3="http://ns.adobe.com/tiff/1.0/"
             xmlns:ns4="http://ns.adobe.com/exif/1.0/"
             xmlns:ns5="http://ns.adobe.com/photoshop/1.0/"
             xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
             ns0:xmptk="Exempi + XMP Core 5.1.2">
      <rdf:RDF>
        <rdf:Description rdf:about="">
          <ns2:CreatorTool>Google</ns2:CreatorTool>
          <ns2:CreateDate>2013-02-09T14:47:53</ns2:CreateDate>
        </rdf:Description>

      .
      .
      .
</ns0:xmpmeta>

Since the UUID data in this case is returned as an lxml ElementTree instance, one can use lxml to access the data. For example, to extract the CreatorTool attribute value, one could do the following

>>> xmp = j.box[3].data
>>> rdf = '{http://www.w3.org/1999/02/22-rdf-syntax-ns#}'
>>> ns2 = '{http://ns.adobe.com/xap/1.0/}'
>>> name = '{0}RDF/{0}Description/{1}CreatorTool'.format(rdf, ns2)
>>> elt = xmp.find(name)
>>> elt
<Element '{http://ns.adobe.com/xap/1.0/#}CreatorTool' at 0xb50684a4>
>>> elt.text
'Google'

But that would be painful. A better solution is to install the Python XMP Toolkit (make sure it is at least version 2.0):

>>> from libxmp import XMPMeta
>>> from libxmp.consts import XMP_NS_XMP as NS_XAP
>>> meta = XMPMeta()
>>> meta.parse_from_str(j.box[3].raw_data.decode('utf-8'))
>>> meta.get_property(NS_XAP, 'CreatorTool')
'Google'

Where the Python XMP Toolkit can really shine, though, is when you are converting an image from another format such as TIFF or JPEG into JPEG 2000. For example, if you were to be converting the TIFF image found at http://photojournal.jpl.nasa.gov/tiff/PIA17145.tif info JPEG 2000:

>>> import skimage.io
>>> image = skimage.io.imread('PIA17145.tif')
>>> from glymur import Jp2k
>>> jp2 = Jp2k('PIA17145.jp2', data=image)

Next you can extract the XMP metadata.

>>> from libxmp import XMPFiles
>>> xf = XMPFiles()
>>> xf.open_file('PIA17145.tif')
>>> xmp = xf.get_xmp()
>>> print(xmp)
<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>
<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="Exempi + XMP Core 5.1.2">
 <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
  <rdf:Description rdf:about=""
    xmlns:tiff="http://ns.adobe.com/tiff/1.0/">
   <tiff:ImageWidth>1016</tiff:ImageWidth>
   <tiff:ImageLength>1016</tiff:ImageLength>
   <tiff:BitsPerSample>
    <rdf:Seq>
     <rdf:li>8</rdf:li>
    </rdf:Seq>
   </tiff:BitsPerSample>
   <tiff:Compression>1</tiff:Compression>
   <tiff:PhotometricInterpretation>1</tiff:PhotometricInterpretation>
   <tiff:SamplesPerPixel>1</tiff:SamplesPerPixel>
   <tiff:PlanarConfiguration>1</tiff:PlanarConfiguration>
   <tiff:ResolutionUnit>2</tiff:ResolutionUnit>
  </rdf:Description>
  <rdf:Description rdf:about=""
    xmlns:dc="http://purl.org/dc/elements/1.1/">
   <dc:description>
    <rdf:Alt>
     <rdf:li xml:lang="x-default">converted PNM file</rdf:li>
    </rdf:Alt>
   </dc:description>
  </rdf:Description>
 </rdf:RDF>
</x:xmpmeta>
<?xpacket end="w"?>

If you are familiar with TIFF, you can verify that there’s no XMP tag in the TIFF file, but the Python XMP Toolkit takes advantage of the TIFF header structure to populate an XMP packet for you. If you were working with a JPEG file with Exif metadata, that information would be included in the XMP packet as well. Now you can append the XMP packet in a UUIDBox. In order to do this, though, you have to know the UUID that signifies XMP data.:

>>> import uuid
>>> xmp_uuid = uuid.UUID('be7acfcb-97a9-42e8-9c71-999491e3afac')
>>> box = glymur.jp2box.UUIDBox(xmp_uuid, str(xmp).encode())
>>> jp2.append(box)
>>> print(jp2.box[-1])
UUID Box (uuid) @ (592316, 1053)
    UUID:  be7acfcb-97a9-42e8-9c71-999491e3afac (XMP)
    UUID Data:
    <ns0:xmpmeta xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:ns0="adobe:ns:meta/" xmlns:ns2="http://ns.adobe.com/tiff/1.0/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" ns0:xmptk="Exempi + XMP Core 5.1.2">
      <rdf:RDF>
        <rdf:Description rdf:about="">
          <ns2:ImageWidth>1016</ns2:ImageWidth>
          <ns2:ImageLength>1016</ns2:ImageLength>
          <ns2:BitsPerSample>
            <rdf:Seq>
              <rdf:li>8</rdf:li>
            </rdf:Seq>
          </ns2:BitsPerSample>
          <ns2:Compression>1</ns2:Compression>
          <ns2:PhotometricInterpretation>1</ns2:PhotometricInterpretation>
          <ns2:SamplesPerPixel>1</ns2:SamplesPerPixel>
          <ns2:PlanarConfiguration>1</ns2:PlanarConfiguration>
          <ns2:ResolutionUnit>2</ns2:ResolutionUnit>
        </rdf:Description>
        <rdf:Description rdf:about="">
          <dc:description>
            <rdf:Alt>
              <rdf:li xml:lang="x-default">converted PNM file</rdf:li>
            </rdf:Alt>
          </dc:description>
        </rdf:Description>
      </rdf:RDF>
    </ns0:xmpmeta>

You can also build up XMP metadata from scratch. For instance, if we try to wrap goodstuff.j2k again:

>>> import glymur
>>> j2kfile = glymur.data.goodstuff()
>>> j2k = glymur.Jp2k(j2kfile)
>>> jp2 = j2k.wrap("goodstuff.jp2")

Now build up the metadata piece-by-piece. It would help to have the XMP standard close at hand:

>>> from libxmp import XMPMeta
>>> from libxmp.consts import XMP_NS_TIFF as NS_TIFF
>>> from libxmp.consts import XMP_NS_DC as NS_DC
>>> xmp = XMPMeta()
>>> ihdr = jp2.box[2].box[0]
>>> xmp.set_property(NS_TIFF, "ImageWidth", str(ihdr.width))
>>> xmp.set_property(NS_TIFF, "ImageHeight", str(ihdr.height))
>>> xmp.set_property(NS_TIFF, "BitsPerSample", '3')
>>> xmp.set_property(NS_DC, "Title", u'Stürm und Drang')
>>> xmp.set_property(NS_DC, "Creator", 'Glymur')

We can then append the XMP in a UUID box just as before:

>>> import uuid
>>> xmp_uuid = uuid.UUID('be7acfcb-97a9-42e8-9c71-999491e3afac')
>>> box = glymur.jp2box.UUIDBox(xmp_uuid, str(xmp).encode())
>>> jp2.append(box)
>>> glymur.set_option('print.codestream', False)
>>> print(jp2)
File:  goodstuff.jp2
JPEG 2000 Signature Box (jP  ) @ (0, 12)
    Signature:  0d0a870a
File Type Box (ftyp) @ (12, 20)
    Brand:  jp2
    Compatibility:  ['jp2 ']
JP2 Header Box (jp2h) @ (32, 45)
    Image Header Box (ihdr) @ (40, 22)
        Size:  [800 480 3]
        Bitdepth:  8
        Signed:  False
        Compression:  wavelet
        Colorspace Unknown:  False
    Colour Specification Box (colr) @ (62, 15)
        Method:  enumerated colorspace
        Precedence:  0
        Colorspace:  sRGB
Contiguous Codestream Box (jp2c) @ (77, 115228)
UUID Box (uuid) @ (115305, 671)
    UUID:  be7acfcb-97a9-42e8-9c71-999491e3afac (XMP)
    UUID Data:
    <ns0:xmpmeta xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:ns0="adobe:ns:meta/" xmlns:ns2="http://ns.adobe.com/tiff/1.0/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" ns0:xmptk="Exempi + XMP Core 5.1.2">
      <rdf:RDF>
        <rdf:Description rdf:about="">
          <ns2:ImageWidth>480</ns2:ImageWidth>
          <ns2:ImageHeight>800</ns2:ImageHeight>
          <ns2:BitsPerSample>3</ns2:BitsPerSample>
        </rdf:Description>
        <rdf:Description rdf:about="">
          <dc:Title>Stürm und Drang</dc:Title>
          <dc:Creator>Glymur</dc:Creator>
        </rdf:Description>
      </rdf:RDF>
    </ns0:xmpmeta>