In this project, you develop an image processing service. Clients send images to a Telegram chatbot, and choose a filter to apply:
Fork this repo, clone it locally and open it in your favorite IDE (PyCharm, VSCode).
You can change the README.md
file content to provide relevant information about your project.
Reference: https://ai.stanford.edu/~syyeung/cvweb/tutorial1.html
If we take a closer look on a digital image, we will notice it comprised of individual pixels, each pixel has its own value. For a grayscale image, each pixel would have an intensity value between 0 and 255, with 0 being black and 255 being white.
A grayscale image, then, can be represented as a matrix of pixel values:
A color image is just a simple extension of this. The colors are constructed from a combination of Red, Green, and Blue (RGB). Instead of one matrix of pixel values, we use 3 different matrix, one for the Red (R) values, one for Green (G), and one Blue (B) values.
As can be seen, each pixel of the image has three channels, represent the red, green, blue values.
Python-wise, a digital grayscale image is essentially a list of lists:
Each element in the image
list is a list represented a row of pixels.
Filtered images are ubiquitous in our social media feeds, news articles, books—everywhere! Image filtering is a technique in image processing that involves modifying or enhancing an image by applying a filter to it. Filters can be used to remove noise, sharpen edges, blur or smooth the image, or highlight specific features or details, among other effects.
Python-wise, image filtering is as simple as manipulate the pixel values.
Under polybot/img_proc.py
, the Img
class is designed for image filtering on grayscale images.
Here is a detailed usage instruction for the class:
Provide the path to the image file as a parameter when creating an instance of the Img
class, for example:
my_img = Img('path/to/image.jpg')
After performing operations on the image, you can save the modified image using the save_img()
method, for example:
my_img.save_img()
This will save the modified grayscale image to a new path with an appended _filtered
suffix, and uses the same file extension.
In this exercise you are required to implement at least one filter from the below list (concat()
, rotate()
, salt_n_pepper()
, segment()
).
You have to implement the filter using Python builtin functionality only. Don't use external packages like numpy, Pillow, openVC, etc.
On every error (e.g. image path doesn't exist, input image is not an RGB) you should raise a RuntimeError
exception.
The concat()
method is meant to concatenate two images together horizontally (side by side).
Implementation instruction for horizontal concatenation:
- Check the dimensions of both images to ensure they are compatible for concatenation. If the dimensions are not compatible (e.g., different heights), raise a
RuntimeError
exception with informative message. - Combine the pixel values of both images to create a new image. For horizontal concatenation, combine each row of the first image with the corresponding row of the second image.
- Store the resulting concatenated image in the
self.data
attribute of the instance.
my_img = Img('path/to/image.jpg')
another_img = Img('path/to/image2.jpg')
my_img.concat(another_img)
my_img.save_img() # concatenated image was saved in 'path/to/image_filtered.jpg'
Note: you can optionally use the direction
argument to implement vertical
concatenation as well.
The salt_n_pepper()
noise method applies a type of image distortion that randomly adds isolated pixels with value of either 255 (maximum white intensity) or 0 (minimum black intensity).
The name "salt and pepper" reflects the appearance of these randomly scattered bright and dark pixels, resembling grains of salt and pepper sprinkled on an image.
Implementation instruction:
- Iterate over the pixels of the image by looping through each row and each pixel value.
- For each pixel in the image:
- Randomly generate a number between 0 and 1.
- If the random number is less than 0.2, set the pixel value to the maximum intensity (255) to represent salt.
- If the random number is greater than 0.8, set the pixel value to the minimum intensity (0) to represent pepper.
- If neither condition is met (the random number is in between 0.2 to 0.8), keep the original pixel value without any modification.
my_img = Img('path/to/image.jpg')
my_img.salt_n_pepper()
my_img.save_img() # noisy image was saved in 'path/to/image_filtered.jpg'
The rotate()
method rotates an image around its center in a clockwise direction.
Implementation remarks:
The resulting rotated image will have its rows become the columns, and the columns will become the rows. The pixels in the rotated image will be repositioned based on a clockwise rotation around the center of the original image. For example, the first row in the original image will become the last column in the rotated image, the second row will become the second-to-last column, and so on.
my_img = Img('path/to/image.jpg')
my_img.rotate()
my_img.rotate() # rotate again for a 180 degrees rotation
my_img.save_img() # rotated image was saved in 'path/to/image_filtered.jpg'
The segment()
method partitions the image into regions where the pixels have similar attributes, so the image is represented in a more simplified manner, and so we can then identify objects and boundaries more easily.
Implementation instruction:
- Iterate over the pixels of the image by looping through each row and each pixel value.
- All pixels with an intensity greater than 100 are replaced with a white pixel (intensity 255) and all others are replaced with a black pixel (intensity 0).
my_img = Img('path/to/image.jpg')
my_img.segment()
my_img.save_img()
The below two filters was already implemented, you can review these functions to get some inspiration of how might a filter implementation look like.
The blur()
method is already implemented. You can control the blurring level blur_level
argument (default is 16).
It blurs the image by replacing the value of each pixel by the average of the 16 pixels around him (or any other value, controlled by the blur_level
argument. The bigger the value, the stronger the blurring level).
my_img = Img('path/to/image.jpg')
my_img.blur() # or my_img.blur(blur_level=32) for stronger blurring effect
my_img.save_img()
The contour()
method is already implemented. It applies a contour effect to the image by calculating the differences between neighbor pixels along each row of the image matrix.
my_img = Img('path/to/image.jpg')
my_img.contour()
my_img.save_img()
Under polybot/test
you'll find unittests for each filter.
For example, to execute the test suite for the concat()
filter, run the below command from the root dir of your repo:
python -m polybot.test.test_concat
An alternative way is to run tests from the Pycharm UI.
- Download and install telegram desktop (you can use your phone app as well).
- Once installed, create your own Telegram Bot by following this section to create a bot. Once you have your telegram token you can move to the next step.
Never commit your telegram token in Git repo, even if the repo is private. For now, we will provide the token as an environment variable to your chat app. Later on in the course we will learn better approaches to store sensitive data.
The Telegram app is a flask-based service that responsible for providing a chat-based interface for users to interact with your image processing functionality. It utilizes the Telegram Bot API to receive user images and respond with processed images.
The code skeleton for the bot app is already given to you under polybot/app.py
.
In order to run the server, you have to provide 2 environment variables:
TELEGRAM_TOKEN
which is your bot token.TELEGRAM_APP_URL
which is your app public URL provided by Ngrok (will be discussed soon).
Implementing bot logic involves running a local Python script that listens for updates from Telegram servers. When a user sends a message to the bot, Telegram servers forward the message to the Python app using a method called webhook (long-polling and websocket are other possible methods which wouldn't be used in this project). The Python app processes the message, executes the desired logic, and may send a response back to Telegram servers, which then delivers the response to the user.
The webhook method consists of simple two steps:
Setting your chat app URL in Telegram Servers:
Once the webhook URL is set, Telegram servers start sending HTTPS POST requests to the specified webhook URL whenever there are updates, such as new messages or events, for the bot.
You've probably noticed that setting localhost
URL as the webhook for a Telegram bot can be problematic because Telegram servers need to access the webhook URL over the internet to send updates.
As localhost
is not accessible externally, Telegram servers won't be able to reach the webhook, and the bot won't receive any updates.
Ngrok can solve this problem by creating a secure tunnel between the local machine (where the bot is running) and a public URL provided by Ngrok. It exposes the local server to the internet, allowing Telegram servers to reach the webhook URL and send updates to the bot.
Sign-up for the Ngrok service (or any another tunneling service to your choice), then install the ngrok
agent as described here.
Authenticate your ngrok agent. You only have to do this once:
ngrok config add-authtoken <your-authtoken>
Since the telegram bot service will be listening on port 8443
, start ngrok by running the following command:
ngrok http 8443
Your bot public URL is the URL specified in the Forwarding
line (e.g. https://16ae-2a06-c701-4501-3a00-ecce-30e9-3e61-3069.ngrok-free.app
).
Don't forget to set the TELEGRAM_APP_URL
env var to your URL.
In the next step you'll finally run your bot app.
Under polybot/bot.py
you are given a class called Bot
. This class implements a simple telegram bot, as follows.
The constructor __init__
receives the token
and telegram_chat_url
arguments.
The constructor creates an instance of the TeleBot
object, which is a pythonic interface to Telegram API. You can use this instance to conveniently communicate with the Telegram servers.
Later, the constructor sets the webhook URL to be the telegram_chat_url
.
The polybot/app.py
is the main app entrypoint. It's nothing but a simple flask webserver that uses a Bot
instance to handle incoming messages, caught in the webhook
endpoint function.
The default behavior of the Bot
class is to "echo" the incoming messages. Try it out!
In bot.py
you are given a class called QuoteBot
which inherits from Bot
.
Upon incoming messages, this bot echoing the message while quoting the original message, unless the user is asking politely not to quote.
In app.py
, change the instantiated instance to the QuoteBot
:
- Bot(TELEGRAM_TOKEN, TELEGRAM_APP_URL)
+ QuoteBot(TELEGRAM_TOKEN, TELEGRAM_APP_URL)
Run this bot and check its behavior.
In bot.py
you are given a class called ImageProcessingBot
which inherits from Bot
, again.
Upon incoming photo messages, this bot downloads the photos and processes them according to the caption
field provided with the message.
The bot will then send the processed image to the user.
A few notes:
- Inside the
ImageProcessingBot
class, overridehandle_message
method and implement the needed functionality. - Remember that by inheriting the
Bot
class, you can use all of its methods (such assend_text
,download_user_photo
,send_photo
...). - Possible
caption
values are:['Blur', 'Contour', 'Rotate', 'Segment', 'Salt and pepper', 'Concat']
.
Note: Your bot should support the Blur
and Contour
filters (those filters have already implemented for you).
Test your bot on real photos and make sure it's functioning properly.
You can test your bot logic locally by:
python -m polybot.test.test_telegram_bot
Or via the Pycharm UI.
Add any functionality you wish to your bot...
- Greet the user.
- Add some informative message when user sends photos without captions or with invalid caption value.
- Add your own filters.
- Extend the functionality of the filters, e.g. allow users to specify "Rotate 2" to rotate the image twice).
Go wild!!!
You don't need to send anything as we already have access to your fork (make sure your forked repo is public). You will be graded by the automated tests running in GitHub action in your fork, make sure you pass them.