klychliiev / innora_test_task

This test task is aimed at the rotation angle detection for rotated document pages

Innora_test_task

Project contents

file	description
invoices_rotated	folder containing rotated images for testing the algorithm
angle_detection.py	script to calculate the rotation angles of input images
image_rotation.py	script to rotate the images from the 'invoices_rotated' folder by the calculated angle
predicted_angles.json	the file generated by the angle_detection.py script. It contains image path as a key and a predicted rotation angle as a value (10 key-value pairs in total)
requirements.txt	libraries used in the project

Run the program

Write the following commands (after the $ sign) in the terminal to run the program:

$ python3 -m venv env
$ source env/bin/activate
$ pip3 install -r requirements.txt
$ python3 angle_detection.py
invoices_rotated/0.png: 7
invoices_rotated/1.png: -2
invoices_rotated/2.png: -9
invoices_rotated/3.png: -7
invoices_rotated/4.png: -1
invoices_rotated/5.png: -6
invoices_rotated/6.png: 5
invoices_rotated/7.png: -2
invoices_rotated/8.png: -4
invoices_rotated/9.png: -1
invoices_rotated/10.png: -4
$ python3 image_rotation.py

The last command writes rotated input images into the 'rotated_images' folder.

Script explanation

The opencv-python library was used to calculate the rotation angle of images. The calculate_angle() function was defined to handle this problem. It takes a path to an image as an argument and processes it in the following way:

1. The image is turned to grayscale;
2. The Canny function function is applied to the grayscale images to detect the edges;
3. The HoughLinesP function detects the lines;
4. The for loop is defined to loop over the detected lines and calculate the angle of each line with respect to the horizontal axis. These angles are stored in the 'angles' list;
5. This function returns the median angle of all the lines in the image thus representing the overall orientation of the image.
   

Example of an image with lines detected using this function:

The rotate_image() function calculates the rotation angle within the [-30,30] degrees range using the predefined calculate_angle() function and rotates each input image by the calculated angle. The rotated images are saved to the 'rotated_images' folder.