Description
0. Set up. Install the latest version of Python, NumPy, PyTorch, Matplotlib, OpenCV. It is
recommended to install Anaconda, as it includes several of these automatically.
You can test your setup by running Python from the command line, then importing modules and
printing their versions.
import numpy as np
import matplotlib
import cv2
print(np.__version__)
print(matplotlib.__version__)
print(cv2.__version__)
There are two popular ways to run Python:
● Command line/ .py file. Edit code with your favorite editor (Visual Studio Code, or
VSCode, is a good choice.) Save to `foo.py` (for example). Then, type `python foo.py` at
the command prompt to run (or use the integrated interpreter via the IDE).
● Jupyter / IPython. Type `jupyter notebook` at the command prompt, and your browser
should automatically open to a page that allows you to work interactively. Alternatively
you could use Google colab to edit/run IPython files. It’s a common practice to use
integrate Google drive with Google colab when your code accesses files (e.g., read or
write images)
1. Image load / save. Load the `original_image.png` image provided. Print the data type of
the image, and the data type of the pixels in the image. Print the image dimensions. Check if
the datatype of the image and a pixel in the image are in uint8 format. Return the data types
and image dimensions.
2. Color channels. Create a new RGB image by setting the green and blue pixel values to zero
everywhere. The resulting image should look red. Note for this and following tasks: The
returned image from the function should be in BGR format for channels and uint8 for pixels
(the native cv2 format). Return the red image.
3. Photographic negative. Create a new RGB image (from the original image) by subtracting each
pixel value from 255. The resulting image should look like a photographic negative. Return the
new negative image.
4. Swap color channels. Create a new RGB image (from the original image) by swapping the red
and blue channels. Return the new image.
5. Foliage detection. Create a new binary image that contains an `ON` pixel wherever the original
input pixel is approximately green, and `OFF` everywhere else. Threshold in such a way that
green is greater than or equal to 50 and others are less than 50. Set `ON` pixels to 255, and
`OFF` pixels to 0. Return the thresholded image.
6. Shift. Translate the original image to the right by 200 pixels, and down by 100 pixels. Fill in the
missing values with zero. Return the shifted image.
7. Rotate. Rotate the original image clockwise by 90 degrees. Return the rotated image.
8. Similarity transform. Write a function to apply a similarity transform to an input image, yielding
an output image of the same size. The function should take scale, rotation angle, and translation
as input (all as floating point values). It should scale, rotate, then translate, in that order. Use
nearest neighbor interpolation for simplicity but be sure to use “inverse mapping” to ensure that
every pixel in the output gets painted. Test this on the original image above by passing the
values scale=2.0, theta=45.0 (degrees), shift=[100,100]. Return the transformed image.
9. Grayscale conversion. Convert the original image into grayscale. Use the formula `gray = (3
* R + 6 * G + 1 * B) / 10`. (Note that this formula has at least two fundamental
problems with it, but it’s commonly used so we’ll ignore those problems for now.) Return the
grayscale image.
The next two tasks will be applied to the binary image `glasses_outline.png`.
10. Moments. Write a function to compute the first- and second-order moments (both standard
and centralized) of a binary image. Apply the function to the image `glasses_outline.png`.
Return the results.
11. Binary image processing. Using the function from the previous problem, write a function to
compute the orientation and eccentricity of `glasses_outline.png`. Orientation should be
clockwise w.r.t. the positive horizontal axis, in degrees. Additionally, draw a “best-fitting ellipse”
on the BGR `glasses_outline.png` in the red channel. Please use the cv2.ellipse
method with a line thickness of 1px. Do not specify a line type or shift. The ellipse should be
drawn in red (0, 0, 255) and should appear on top of the white (255, 255, 255) outline
of the glasses. Return the orientation, eccentricity, and modified image with the ellipse.
