Computer Vision in Python
Mahotas is a computer vision library for Python. It contains traditional image processing functionality such as filtering and morphological operations as well as more modern computer vision functions for feature computation, including interest point detection and local descriptors. The interface is in Python, a dynamic programming language, which is very appropriate for fast development, but the algorithms are implemented in C++ and are tuned for speed. The library is designed to fit in with the scientific software ecosystem in this language and can leverage the existing infrastructure developed in that language. Mahotas is released under a liberal open source license (MIT License) and is available from (http://github.com/luispedro/mahotas) and from the Python Package Index (http://pypi.python.org/pypi/mahotas).READ FULL TEXT VIEW PDF
scikit-image is an image processing library that implements algorithms a...
In this work, user's emotion using its facial expressions will be detect...
There are high technological and software demands associated with conduc...
This paper introduces GIMP-ML, a set of Python plugins for the widely po...
While the Python language has extension frameworks such as NumPy and Sci...
In this project, we have used the computer vision algorithm SSD (Single ...
This work presents Kornia, an open source computer vision library built ...
Computer Vision in Python
Read images to numpy arrays
Read image data – TIFF, PNG, JPG, WebP, PVR Textures, and others – into numpy arrays
Computer Vision in Python
Computer Vision in Python
Mahotas is a computer vision library for Python. It contains traditional image processing functionality such as filtering and morphological operations as well as more modern computer vision functions for feature computation, including interest point detection and local descriptors.
The interface is in Python, a dynamic programming language, which is very appropriate for fast development, but the algorithms are implemented in C++ and are tuned for speed. The library is designed to fit in with the scientific software ecosystem in this language and can leverage the existing infrastructure developed in that language.
Mahotas is released under a liberal open source license (MIT License) and is available from http://github.com/luispedro/mahotas and from the Python Package Index (http://pypi.python.org/pypi/mahotas).
Keywords: computer vision, image processing.
Mahotas is a computer vision library for the Python Programming Language (versions 2.5 and up, including version 3 and up). It operates on numpy arrays (van der Walt et al., 2011). Therefore, it uses all the infrastructure built by that project for storing information and performing basic manipulations and computations. In particular, unlike libraries written in the C Language or in Java (Marcel and Rodriguez, 2010), Mahotas does not need to define a new image data structure, but uses the numpy array structure. Many basic manipulation functionality that would otherwise be part of a computer vision library are handled by numpy. For example, computing averages and other simple statistics, handling multi-channel images, converting between types (integer and floating point images are supported by mahotas) can all be performed with numpy builtin functionality. For the user, this has the additional advantage that they do not need to learn yet another set of functions.
It contains over 100 functions with functionality ranging from traditional image filtering and morphological operations to more modern wavelet decompositions and local feature computations. Additionally, by integrating into the Python numeric ecosystem, users can use other packages in a seamless way. In particular, mahotas does not implement any machine learning functionality, but rather advises the user to use another, specialised package, such as scikits-learn or milk.
Python is a natural “glue” language: it is easy to use state-of-the-art libraries written in multiple languages (Oliphant, 2007). Mahotas itself is a mix of high-level Python and low-level C++. This achieves a good balance between speed and ease of implementation.
Version 1.0 of mahotas has been released recently and this is now a mature, well-tested package (the first versions were made available over 4 years ago, although the package was not named mahotas then). 111Note for reviewers: the version currently available (0.9.6) implements all functionality described in this manuscript. I will release it as version 1.0 when this manuscript is accepted to coincide with publication. Naturally, any bugs that are found and reported in the meanwhile will be addressed. Mahotas runs and is used on different versions of Unix (including Linux, SunOS, and FreeBSD), Mac OS X, and Windows.
The interface is a procedural interface, with no global state. All functions work independently of each other (there is code sharing at the implementation level, but this is hidden from the user).
The main functionality is grouped into the following categories:
Speeded-up Robust Features (Bay et al., 2008). This includes both keypoint detection and descriptor computation.
Haar and Daubechies wavelets. Forward and inverse transforms are supported.
Erosion and dilation, as well as some more complex operations built on these. There are both binary and grayscale implementations of these operators.
seeded watershed and distance map transforms (Felzenszwalb and Huttenlocher, 2004).
Gaussian filtering, edge finding, and general convolutions.
convex hull, polygon drawing.
Numpy arrays contain data of a specific type, such unsigned 8 bit integer or floating point numbers. While natural colour images are typically 8 bits, scientific data is often larger and processing can result in floating point images. Mahotas works on all datatypes. This is performed without any extra memory copies. Mahotas is heavily optimised for both speed and memory usage (it can be used with very large arrays).
There are a few interface conventions which apply to many functions. When meaningful, a structuring element is used to define neighbourhoods or adjacency relationships (morphological functions, in particular, use this convention). Generally, the default is to use a cross as the default if no structuring filter is given.
When a new image is to be returned, functions take an argument named out where the output will be stored. This argument is often much more restricted in type. In particular, it must be a contiguous array.222Numpy supports non-contiguous arrays, which are most often slices into other, larger, contiguous arrays (e.g., given a contiguous array, one can build a non-contiguous array by taking every other row). Since this is a performance feature (its purpose is to avoid extra memory allocation), it is natural that the interface is less flexible (accessing a contiguous array is much more efficient than a non-contiguous one).
Code for this and other examples is present in the mahotas source distribution under the demos/ directory. In this example, we load an image, find SURF interest points, and compute descriptors.
We start by importing the necessary packages, including numpy and mahotas. We also use milk, to demonstrate how the mahotas output can integrate with a machine learning package.
import numpy as np import mahotas from mahotas.features import surf import milk
The first step is to load the image and convert to 8 bit numbers. In this case, the conversion is done using standard numpy methods, namely astype:
f = mahotas.imread(’luispedro.jpg’, as_grey=True) f = f.astype(np.uint8)
We can now compute SURF interest points and descriptors. spoints = surf.surf(f, 4, 6, 2)
The surf.surf function returns both the descriptors and their meta data. We use numpy operations to retain only the descriptors (the meta data is in the first five positions):
descrs = spoints[:,6:]
Using milk, we cluster the descriptors into five groups:
values, _= milk.kmeans(descrs, 5)
Finally, we can show the points in different colours. colors = np.array( [ 255, 25, 1], [203, 77, 37], [151, 129, 56], [ 99, 181, 52], [ 47, 233, 5]]) f2 = surf.show_surf(f, spoints[:64], values, colors)
from matplotlib import pyplot as plt plt.subplot(1,2,1) plt.imshow(f) plt.subplot(1,2,2) plt.imshow(f2)
The easy interaction with matplotlib is another way in which we benefit from the numpy-based ecosystem as mahotas does not need to support interacting with a graphical system to display images.
Mahotas is mostly written in C++, but this is completely hidden from the user as there are hand-written Python wrappers for all functions. Automatically generated wrappers inevitably lead to worse error messages and are less flexible.
The main reason that mahotas is implemented in C++ (and not in C, which is the language of the Python interpreter) is to use templates. Almost C++ functionality is split across 2 functions:
A py_function which uses the Python C API to get arguments and check them.
A template function<dtype> which works for the type dtype performing the actual operation.
So, for example, this is how erode is implemented.333This is the generic version of the code. Erode, like a few other functions, has two versions, a fast version, limited to two-dimensional, contiguous, images and a generic one presented here. The selection between the two implementations is done automatically for the user. py_erode consists mostly of boiler-plate code:
PyObject* py_erode(PyObject* self, PyObject* args) PyArrayObject* array; PyArrayObject* Bc; PyArrayObject* output; if (!PyArg_ParseTuple(args, ”OOO”, &array, &Bc, &output) —— !numpy::are_arrays(array, Bc, output) —— !numpy::same_shape(array, output) —— !numpy::equiv_typenums(array, Bc, output) —— PyArray_NDIM(array) != PyArray_NDIM(Bc) ) PyErr_SetString(PyExc_RuntimeError, TypeErrorMsg); return NULL; holdref r_o(output);
#define HANDLE(type) erode¡type¿(numpy::aligned_array¡type¿(output), numpy::aligned_array¡type¿(array), numpy::aligned_array¡type¿(Bc)); SAFE_SWITCH_ON_INTEGER_TYPES_OF(array); #undef HANDLE …
This functions retrieves the arguments, performs some sanity checks, performs a bit of initialisation, and finally, switches in the input type with the help of the SAFE_SWITCH_ON_INTEGER_TYPES macro, which call the right specialisation of the template that does the actual work. In this example erode implements erosion:
template¡typename T¿ void erode(numpy::aligned_array¡T¿ res, numpy::aligned_array¡T¿ array, numpy::aligned_array¡T¿ Bc) gil_release nogil; const int N = res.size(); typename numpy::aligned_array¡T¿::iterator iter = array.begin(); filter_iterator¡T¿ filter(array.raw_array(), Bc.raw_array(), ExtendNearest, is_bool(T())); const int N2 = filter.size(); T* rpos = res.data();
for (int i = 0; i != N; ++i, ++rpos, filter.iterate_both(iter)) T value = std::numeric_limits¡T¿::max(); for (int j = 0; j != N2; ++j) T arr_val = T(); filter.retrieve(iter, j, arr_val); value = std::min¡T¿(value, erode_sub(arr_val, filter[j])); *rpos = value;
The template machinery makes the functions that use it very simple and easy to read. The only downside is that there is some expansion of code size when the compiler instantiates the function for the several integer and floating point types. Given the small size of these functions, the total size of the compiled library is reasonable (circa 6MiB on an Intel-based 64 bit system for the whole library).
In the snippet above, you can see some other C++ machinery:
This is a “resource-acquisition is object initialisation” (raii)444Raii is a design pattern in C++, or other languages with scope linked deterministic object destruction, such as D, where a resource is represented by an object, whose constructor acquires it and whose destructor releases it. This guarantees that the object is correctly released even if the scope is left through an exception (Stroustrup, 1994). object that releases the Python global interpreter lock (gil)555In the CPython interpreter, the most commonly used implementation of Python, there is a global lock for many Python related functionality, which limits parallelism. in its constructor and gets it back in its destructor. Normally, the template function will release the gil after the Python-specific code is done. This allows several mahotas functions to run concurrently.
This is a thin wrapper around PyArrayObject, the raw numpy data type, which has iterators which resemble the C++ standard library. It also handles type-casting internally, making the code type-safer. This is also a raii object in terms of managing Python reference counts. In mahotas debug builds, this object additionally adds several checks to all the memory acesses.
This was adapted from code in the scipy.ndimage packages and it is useful to iterate over an image and use a centered filter around each pixel (it keeps track of all of the boundary conditions).
The inner loop is as direct an implementation of erosion as one would wish for: for each pixel in the image, look at its neighbours, subtract the filter value, and compute the minimum of this operation.
|median filter (2)||226.9||NA||2034.0||NA|
|median filter (10)||2810.9||NA||1877.1||NA|
Table 1 shows timings for different operations. These were normalized to multiples of the time it takes to go over the image and find its maximum pixel value (this was done using numpy.max(image)). The measurements shown were obtained on an Intel 64 bit system, running Ubuntu Linux. However, due to the normalization, measurements obtained on another system (Intel 32 bits running Mac OS) were qualitatively similar.
The comparison is against Pymorph (Dougherty and Lotufo, 2003), which is a pure Python implementation of some of the same functions; scikits-image, which is a similar project to mahotas, but with a heavier emphasis on the use of Cython (Behnel et al., 2011); and OpenCV, which is a C++ library with automatically generated Python wrappers.
OpenCV is the fastest library, but this comes at the cost of some flexibility. Arguments to its functions must be of the exact expected type and it is possible to crash the interpreter if types do match the expected type (in the other libraries, including mahotas, all types are checked and an exception is generated which can be caught by user code). This is particularly relevant for interactive use as the user is often exploring and is willing to pay the speed cost of a few extra type checks and conversions to avoid a hard-crash.
Among the other libraries, mahotas is the fastest. Pymorph, even though it is implemented in Python only, intelligently uses arithmetic operations for morphological operation and can be very fast. However, for more complex methods, such as watershed; its pure Python approach is very inefficient. The one exception is that median filtering with a large structuring element is faster in scikit-image. In fact, that library uses an algorithm with better asymptotic behavior for this operation.
In keeping with the philosophy of blending in with the ecosystem, Mahotas uses the standard Python build machinery and distribution channels. Building and installing from source code is done using
python setup.py install
Alternatively, Python based package managers (such as easy_install or pip) can be used (mahotas works well with these systems).
Mahotas includes a complete automated suite of unit tests, which tests all functionality and include several regression tests. There are no known bugs in version 1.0. In fact, no releases have ever been performed with known bugs. Naturally, bugs were, occasionally, discovered in released versions, but corrected before the next release.
The development is completely open-source and development versions are available. Many users have submitted bug reports and fixes.
Mahotas runs and is used on different versions of Unix (including Linux, SunOS, and FreeBSD), Mac OS X, and Windows.666Christoph Gohlke has been instrumental in providing Windows packages as well as several fixes for that platform.
Mahotas works in Python (minimal version is 2.5, but mahotas works with all more recent versions, including version in the Python 3 series).
Additional system requirements
None at runtime. Compilation from source requires a C++ compiler.
It requires numpy to be present and installed.
List of contributors
Luis Pedro Coelho (Carnegie Mellon University and Instituto de Medicina Molecular), Zachary Pincus (Stanford University), Peter J. Verveer (European Molecular Biology Laboratory), Davis King (Northrop Grumman ES), Robert Webb (Carnegie Mellon University), Matthew Goodman (University of Texas at Austin), K.-Michael Aye (University of Bern), Rita Simões (University of Twente), Joe Kington (University of Wisconsin), Christoph Gohlke (University of California, Irvine), Lukas Bossard (ETH Zurich), and Sandro Knauss (University of Bremen).
Date published: Since 2010 as mahotas. Some of the code had been previously made available under other names.
Originally, this code was developed in the context of cellular image analysis. However, none of the functionality is specific to this context and many computer vision pipelines can make use of it.
This package (and earlier versions of it) have been used by myself (Coelho et al., 2009, 2010a) and close collaborators in several publications (Cho et al., 2012). Other groups have used in published work, both in cell image analysis (Mashburn et al., 2012) and in other areas (Machálek and Olševičová, 2013).
Python is an excellent language for scientific programming because of the inherent properties of the language and because of the infrastructure that has been built around the numpy project. Mahotas works in this environment to provide the user with image analysis and computer vision functionality.
Mahotas does not include machine learning related functionality, such as -means clustering or classification methods. This is the result of an explicit design decision. Specialised machine learning packages for Python already exist (Pedregosa et al., 2011; Demšar et al., 2004; Schaul et al., 2010; Sonnenburg et al., 2010). A good classification system can benefit both computer vision users and others. As these projects all use Numpy arrays as their data types, it is easy to use functionality from the different project seamlessly (no copying of data is necessary).
Mahotas is implemented in C++, as the standard Python interpreter is too slow for a direct Python implementation. However, all of the Python interface code is hand-written, as opposed to using automatic interface generators like Swig (Beazley, 2003). This is more work initially, but the end result is of much higher quality, especially when it comes to giving useful error messages (e.g., when a type mismatch occurs, an automatic system will often be forced to resort to a generic message as it does not have any knowledge of what the arguments mean besides their automatically inferred types).
Mahotas has been available in the Python Package Index since April 2010 and has been downloaded over 40,000 times. This does not include any downloads from other sources. Mahotas includes a full test suite. There are no known bugs.
Mahotas includes code ported and incorporated from other projects. Initially, it was used in reproducing the functionality in the Subcellular Location Image Classifier (SLIC) tool from Robert F. Murphy’s Lab(Zhao and Murphy, 2006) and the initial versions of mahotas were designed explicitly to support that functionality. The surf implementation is a port from the code from dlib,777Dlib’s webpage is at http://dlib.net. a very good C++ library by Davis King. I also gleaned some insight into the implementation of these features from Christopher Evan’s OpenSURF library and its documentation (Evans, 2009).888OpenSURF is available at http://www.chrisevansdev.com/computer-vision-opensurf.html, where several documents describe details of the implementation. The code which interfaces with the FreeImage library, was written by Zachary Pincus and some of the support code was written by Peter J. Verveer for the scipy.ndimage project. All of these contributions were integrated while respecting the software licenses under which the original code had been released. Robert Webb, a summer student at Carnegie Mellon University, worked with me on the initial local binary patterns implementation. Finally, I thank the several users who have reported bugs, submitted fixes, and participated on the project mailing list.
Stéfan van der Walt and Andreas Müller offered helpful comments on a draft version of this manuscript.
Funding: I was supported in my work by the Fundação para a Ciência e Tecnologia (grants SFRH/BD/37535/2007 and PTDC/SAU-GMG/115652/2008), by NIH grant GM078622 (to Robert F. Murphy), by a grant from the Scaife Foundation, by the HHMI Interfaces Initiative, and by a grant from the Siebel Scholars Foundation.
Torchvision the machine-vision package of torch.In Proceedings of the international conference on Multimedia, MM ’10, pages 1485–1488, New York, NY, USA. ACM.