top bar

Using the Micro-Manager python library

MMCorePy is a wrapper that allows you to control microscope hardware from python interactive session or script. It's support Windows, Mac and Linux.

Micromanager's main parts:

  • CMMCore - basic module, written in C++. Script languages like python just wrap it by swig.
  • Device adapters - various libraries that allow support for various hardware. If you want to built one and extend MM devise support, follow this guide.
  • MMCorePy - python wrapper. MM build scripts are support both python 2 and 3, but windows version still ships with python 2 bindings only.
  • MMCoreJ - java wrapper
  • MMStudio - Micromanager GUI (technically it is ImageJ plugin).

Environment setup

You must install python2 and numpy. Windows users may prefer using an python distribution instead manual separate installation.

Manual

  • python 2.7.x (python2 is default for windows now)
  • numpy 1.7.x Micromanager represent imaging data as multidimensional numpy arrays.

Using python distributions

It's convenient to install a distribution which includes Python, numpy, scientific libraries, GUI frameworks and IDEs. All distributions have a free version, some of them have extended paid version, and you can request free academic license.

Useful libraries

  • Scipy - scientific algorithms, multidimensional image processing toolbox.
  • Matplotlib - fastest way to show your image data.
  • Opencv - computer vision and image processing library. Sometimes faster than scipy.
  • Pillow - very basic image processing. Scipy uses it for image loading and writing.
  • Scikit-image - "pythonic" scientific-oriented image processing algorithms collection.
  • IPython - improved interactive python environment

Micromanager installation

Windows & Mac

Download and install Micro-Manager on your computer. Add Micromanager installation folder to PYTHONPATH (i.e. "C:\Program Files\Micro-Manager-1.4", it should contain MMCorePy.py and _MMCorePy.pyd files). Create variable if it not exist. At now you can import MMCorePy without an error.

Linux

MM package from your distribution repository, in most cases, ships with CMMCore, MMCorePy (python 2 or 3 wrapper) and MMCoreJ, but without GUI. Python and Numpy would be installed by your package manager as dependency. In normal way, you don't need changing any system variable.

Using Python API

Familiarize yourself with Micro-Manager and learn how to connect it to your hardware by MMStudio GUI.

First steps

Note
Next code snippets aims to be most generic. We use numpy and matplotlib "as is" from pure python interactive shell, but it's convenient to use IPython with nice autocompletion capabilities.

Start python interactive session. Import `MMCorePy` and make sure everything is working properly.

   >>> import MMCorePy
   >>> mmc = MMCorePy.CMMCore()  # Instance micromanager core
   >>> mmc.getVersionInfo()
   'MMCore version 2.3.2'
   >>> mmc.getAPIVersionInfo()
   'Device API version 59, Module API version 10'

We just get some basic information about current Micromanager installation. If there an `ImportError`, check your PYTHONPATH variable. If output is too verbose, run mmc.enableStderrLog(False); mmc.enableDebugLog(False).

Device loading

Note
You can get all needed parameter's names from Micromanager configuration file, generated by MMStudio.

Let's take step closer to hardware. Micromanager have couple of dummy devices, suitable for learning purposes. Load DemoCamera:

   # Demo camera example, continuation of previous listing
   >>> mmc.loadDevice('Camera', 'DemoCamera', 'DCam')
   >>> mmc.initializeAllDevices()
   >>> mmc.setCameraDevice('Camera')

Property discovery

Every device has properties - settings that let you control the device more precisely. Default values should be fine, but if you need something sophisticated, this example help you figure out how to explore it.

Snapping single image

Images returned as numpy array by calls to an instance of the pythonized Micro-Manager CMMCore class. The array dtype depends on property named PixelType (see below).

Grayscale

   >>> mmc.snapImage()
   >>> img = mmc.getImage()  # img - it's just numpy array
   >>> img
   array([[12, 12, 13, ..., 11, 12, 12],
          [12, 12, 13, ..., 11, 12, 12],
          [12, 13, 13, ..., 12, 12, 12],
          ...,
          [22, 22, 22, ..., 22, 22, 22],
          [22, 22, 22, ..., 22, 22, 22],
          [22, 22, 22, ..., 22, 22, 22]], dtype=uint8)

DemoCamera snaps grayscale 8-bit image, by default. It presented as two-dimensional numpy array. Let's show image data with matplotlib.

   >>> import matplotlib.pyplot as plt
   >>> plt.imshow(img, cmap='gray')
   >>> plt.show()  # And window will appear

Color

Of course, color image is more suitable for optical microscopy purposes. So take one, if your camera support it:

   >>> mmc.setProperty('Camera', 'PixelType', '32bitRGB')  # Change pixel type
   >>> rgb32 = mmc.getImage()
   >>> rgb32
   array([[1250067, 1250067, 1315860, ..., 1250067, 1250067, 1250067],
          [1250067, 1315603, 1315860, ..., 1250067, 1250067, 1250067],
          [1250067, 1315859, 1315860, ..., 1250067, 1250067, 1250067],
          ...,
          [1246483, 1246483, 1246483, ..., 1181204, 1246740, 1246484],
          [1246483, 1246483, 1246483, ..., 1246740, 1246740, 1246483],
          [1246483, 1246483, 1312019, ..., 1246740, 1246740, 1246483]], dtype=uint32)

Interesting output isn't it? We expect something like 3-dimensional RGB array, but get bunch of 32-bit uints in 2-D shape.

Numpy array

Now we should look at RGB32 pixel data structure. Every pixel has 32-bit depth and contain 4 values for blue, green, red and blank channel. Blank channel is more technical peculiarity, than necessity.

   low memory address    ---->      high memory address
   | pixel | pixel | pixel | pixel | pixel | pixel |...
   |-------|-------|-------|-------|-------|-------|...
   |B|G|R|A|B|G|R|A|B|G|R|A|B|G|R|A|B|G|R|A|B|G|R|A|...
   http://avisynth.nl/index.php/RGB32

Let's numpy handle that.

   >>> import numpy as np
   >>> rgb32.shape
   (512, 512)
   >>> rgb = rgb32.view(dtype=np.uint8).reshape(
           rgb32.shape[0], rgb32.shape[1], 4)[...,2::-1]
   >>> rgb.shape
   (512, 512, 3)
   >>> rgb.dtype
   dtype('uint8')

It is a fastest way to get pixel data as RGB array without copying. There is no conversion - just creating new view to same data. Now you can process image with scipy or scikits-image. Note, that opencv uses BGR order (replace slice to [..., :3] for that).

Continuous acquisition

Warning
Don't run this code directly. It's a partial sample.
   mmc.startContinuousSequenceAcquisition(1)
   while True:
       if mmc.getRemainingImageCount() > 0:
           frame = mmc.getLastImage()
           # or frame = mmc.popNextImage()

Code examples

A longer example script, MMCoreWrapDemo.py, is available in the Micro-Manager root directory.

Also check out micromanager-samples repo (live video acquisition, property discovery etc).

Further reading

Written by Eugene Dvoretsky -- Radioxoma (talk) 09:19, 14 June 2014 (PDT)

© Micro-Manager : Vale Lab, UCSF 2006-2011 | All Rights Reserved | Contact