Introduction#

Rationale#

The Machine Vision Toolbox (MVTB) brings professional-grade vision algorithms to your fingertips with a Pythonic API. Built on a “best-of-breed” foundation—NumPy, SciPy, OpenCV, and Open3D —- it bridges the gap between raw pixel manipulation and high-level spatial reasoning.

While the Python ecosystem offers powerful individual tools, using them in concert is often complex. Libraries like OpenCV, scikit-image and Open3D provide excellent algorithms, but they frequently differ in API style, coordinate conventions, and expected data types. MVTB unifies these capabilities into a consistent, object-oriented framework. By treating images and cameras as first-class objects rather than just raw arrays, the Toolbox allows you to focus on the geometry and logic of your vision system rather than the boilerplate of library integration.

For example, to read an image using OpenCV, smooth it, and display it is:

import cv2
import numpy

# read image
src = cv2.imread(".../flowers1.png", cv2.IMREAD_UNCHANGED)

# apply Gaussian blur on src image
dst = cv2.GaussianBlur(src, (5,5), cv2.BORDER_DEFAULT)

# display input and output image
cv2.imshow("Gaussian Smoothing",numpy.hstack((src, dst)))
cv2.waitKey(0) # waits until a key is pressed
cv2.destroyAllWindows() # destroys the window showing image

Using this toolbox we would write instead:

from machinevisiontoolbox import Image

img = Image.Read("flowers1.png") # read the image
smooth = img.smooth(hw=2)  # apply a Gaussian blur
smooth.disp(block=True)  # display and block until window dismissed

or even:

from machinevisiontoolbox import Image

img = Image.Read("flowers1.png").smooth(hw=2).disp(block=True)

which exploits the power of Python’s method chaining – allowing a processing pipeline to be expressed in a single line of very readable code.

While the merits (or demerits) of these different approaches is subjective, you get the idea that the Toolbox allows succinct coding without the need for lots of OpenCV flags like cv2.IMREAD_UNCHANGED in the example above.

In summary, the Machine Vision Toolbox for Python (MVTB-P):

  • provides many functions that are useful in machine vision and vision-based control.

  • provides a simple, yet powerful and consistent, object-oriented wrapper of OpenCV functions. It supports operator overloading and handles the gnarly details of OpenCV-like conversion to/from float32 and the BGR color order.

  • leverages the power of NumPy and OpenCV, and inherits their efficiency, portability and maturity.

  • has similar, but not identical, functionality to the older Machine Vision Toolbox for MATLAB.

  • includes over 100 functions such as image file reading and writing, acquisition, display, filtering, blob, point and line feature extraction, mathematical morphology, homographies, visual Jacobians, camera calibration and color space conversion. With input from a web camera and output to a robot (not provided) it would be possible to implement a visual servo system entirely in Python.

  • includes functionality spanning photometry, photogrammetry, colorimetry; while also being sufficient to support the book Robotics, Vision & Control.

Image objects#

The key element of the Toolbox is the Image class. This sections provides some examples, but full details are given in The Image object. The remainder of this section provides a brief overview of the key features of the Image class with examples.

Firstly, there are lots of ways to create an image. We can read an image from a file:

img = Image.Read("street.png")

or create it from code:

img = Image.Zeros(100, dtype="uint8")

Under the hood the Image object contains some image parameters, a lot of methods, and a reference to a 2D or 3D NumPy ndarray containing the pixel data.

Image object methods generally consider pixel coordinates with the horizontal coordinate first and the vertical coordinate second – consistent with the way we write about algorithms but the opposite to the way that NumPy indexes an array.

An image object has a lot of useful attributes that describe the image, including:

  • img.width, the width of the image in pixels

  • img.height, the height of the image in pixels

  • img.size, the size of the image (width, height) in pixels

  • img.nplanes, the number of planes in the image

as well as a number of useful predicates including:

  • img.iscolor, is the image multichannel?

  • img.ismono, is the image single channel?

  • img.isfloat, does the image have floating point pixels?

Accessing the pixel array#

We can access the array of pixel values by the array attribute:

np.mean(img.array)

We can slice the image using the same syntax as a NumPy array:

img[10:20, 30:40]

but only for reading, not for assignment. The result is another Image object.

Multi-plane images#

Color images are handled a bit more sensibly than raw OpenCV. A multi-channel or multi-plane image is a NumPy ndarray with an arbitrary number of planes and a dictionary that maps channel names to an integer index. For instance, to create multi-plane images we can write any of the following:

img = Image.Zeros(100, colororder="RGB")
img = Image.Zeros(100, colororder="XYZ")
img = Image.Zeros(100, colororder="red:green:blue")
img = Image.Zeros(100, colororder="PQRST")  # 5 channel image

which create 100x100 images with 3, 3, 3 and 5 planes respectively, with all pixel values set to zero. Rather than have the meaning of the plane implicit (ie. plane 0 is red), it is explicit, for example:

img.plane("R")
img.plane("Y")
img.plane("blue")

A more common example is to read a color image:

img = Image.Read("flowers1.png")
img.red().disp()  # display the red plane of the image, whether RGB or BGR format
img.colorspace("hsv").plane("h").disp()  # display the hue plane of an HSV image

Image iterators#

Frequently we want to use images that form a seqeuence – consecutive frames from a camera or a video file, a web camera, image files in a folder or zip file. Rather than build this capability into the Image object we provide a number of iterator objects:

for img in FileArchive("holidaypix.zip"):
        # process the image

Getting started#

Using pip#

Install a snapshot from PyPI:

$ pip install machinevision-toolbox-python

From GitHub source#

Install the current code base from GitHub and pip install a link to that cloned copy:

$ git clone https://github.com/petercorke/machinevision-toolbox-python.git
$ cd machinevision-toolbox-python
$ pip install -e .

Examples#

Binary blobs#

We load a binary image of two sharks and find the blobs in the image. We then display the image with the blobs marked by bounding boxes and centroids.

import machinevisiontoolbox as mvtb
import matplotlib.pyplot as plt
im = mvtb.Image("shark2.png")   # read a binary image of two sharks
fig = im.disp();   # display it with interactive viewing tool
f = im.blobs()  # find all the white blobs
print(f)

which will display:

┌───┬────────┬──────────────┬──────────┬───────┬───────┬─────────────┬────────┬────────┐
│id │ parent │     centroid │     area │ touch │ perim │ circularity │ orient │ aspect │
├───┼────────┼──────────────┼──────────┼───────┼───────┼─────────────┼────────┼────────┤
│ 0 │     -1 │ 371.2, 355.2 │ 7.59e+03 │ False │ 557.6 │       0.341 │  82.9° │  0.976 │
│ 1 │     -1 │ 171.2, 155.2 │ 7.59e+03 │ False │ 557.6 │       0.341 │  82.9° │  0.976 │
└───┴────────┴──────────────┴──────────┴───────┴───────┴─────────────┴────────┴────────┘

f.plot_box(fig, color='g')  # put a green bounding box on each blob
f.plot_centroid(fig, 'o', color='y')  # put a circle+cross on the centroid of each blob
f.plot_centroid(fig, 'x', color='y')
plt.show(block=True)  # display the result
Binary image showing bounding boxes and centroids

Binary blob hierarchy#

We load a binary image with nested objects

im = mvtb.Image("multiblobs.png")
im.disp()
Binary image showing bounding boxes and centroids 
f  = im.blobs()
print(f)

which will display:

┌───┬────────┬───────────────┬──────────┬───────┬────────┬─────────────┬────────┬────────┐
│id │ parent │      centroid │     area │ touch │  perim │ circularity │ orient │ aspect │
├───┼────────┼───────────────┼──────────┼───────┼────────┼─────────────┼────────┼────────┤
│ 0 │      1 │  898.8, 725.3 │ 1.65e+05 │ False │ 2220.0 │       0.467 │  86.7° │  0.754 │
│ 1 │      2 │ 1025.0, 813.7 │ 1.06e+05 │ False │ 1387.9 │       0.769 │ -88.9° │  0.739 │
│ 2 │     -1 │  938.1, 855.2 │ 1.72e+04 │ False │  490.7 │       1.001 │  88.7° │  0.862 │
│ 3 │     -1 │  988.1, 697.2 │ 1.21e+04 │ False │  412.5 │       0.994 │ -87.8° │  0.809 │
│ 4 │     -1 │  846.0, 511.7 │ 1.75e+04 │ False │  496.9 │       0.992 │ -90.0° │  0.778 │
│ 5 │      6 │  291.7, 377.8 │  1.7e+05 │ False │ 1712.6 │       0.810 │ -85.3° │  0.767 │
│ 6 │     -1 │  312.7, 472.1 │ 1.75e+04 │ False │  495.5 │       0.997 │ -89.9° │  0.777 │
│ 7 │     -1 │  241.9, 245.0 │ 1.75e+04 │ False │  496.9 │       0.992 │ -90.0° │  0.777 │
│ 8 │      9 │ 1228.0, 254.3 │ 8.14e+04 │ False │ 1215.2 │       0.771 │ -77.2° │  0.713 │
│ 9 │     -1 │ 1225.2, 220.0 │ 1.75e+04 │ False │  496.9 │       0.992 │ -90.0° │  0.777 │
└───┴────────┴───────────────┴──────────┴───────┴────────┴─────────────┴────────┴────────┘

We can display a label image, where the value of each pixel is the label of the blob that the pixel belongs to

out = f.labelImage(im)
out.stats
out.disp(block=True, colormap="jet", cbar=True, vrange=[0,len(f)-1])

and request the blob label image which we then display

Binary image showing bounding boxes and centroids

Camera modelling#

>>> cam = mvtb.CentralCamera(f=0.015, rho=10e-6, imagesize=[1280, 1024], pp=[640, 512], name="mycamera")
>>> print(cam)
                        Name: mycamera [CentralCamera]
        focal length: (array([0.015]), array([0.015]))
          pixel size: 1e-05 x 1e-05
        principal pt: (640.0, 512.0)
          image size: 1280.0 x 1024.0
        focal length: (array([0.015]), array([0.015]))
                        pose: t = 0, 0, 0; rpy/zyx = 0°, 0°, 0°

and its intrinsic parameters are

>>> print(cam.K)
        [[1.50e+03 0.00e+00 6.40e+02]
        [0.00e+00 1.50e+03 5.12e+02]
        [0.00e+00 0.00e+00 1.00e+00]]

We can define an arbitrary point in the world

>>> P = [0.3, 0.4, 3.0]

and then project it into the camera

>>> p = cam.project(P)
print(p)
        [790. 712.]

which is the corresponding coordinate in pixels. If we shift the camera slightly the image plane coordinate will also change

>>> p = cam.project(P, T=SE3(0.1, 0, 0) )
>>> print(p)
[740. 712.]

We can define an edge-based cube model and project it into the camera’s image plane

>>> X, Y, Z = mkcube(0.2, pose=SE3(0, 0, 1), edge=True)
>>> cam.mesh(X, Y, Z)
Perspective camera view

Color space#

Plot the CIE chromaticity space

>>> showcolorspace("xy")
CIE chromaticity space

Load the spectrum of sunlight at the Earth’s surface and compute the CIE xy chromaticity coordinates

>>> nm = 1e-9
>>> lam = np.linspace(400, 701, 5) * nm # visible light
>>> sun_at_ground = loadspectrum(lam, 'solar')
>>> xy = lambda2xy(lambda, sun_at_ground)
>>> print(xy)
        [[0.33272798 0.3454013 ]]
>>> print(colorname(xy, 'xy'))
        khaki