differences-between-two-images (or multiple images)

Detect and visualize differences between images with OpenCV and the SSIM method. The images are compared with the first provided image.

Versions

The latest version of this project is currently 2.0.1.

Note: to navigate through the different versions, use e.g. git checkout tags/1.0.1.

Dependencies

To manage the dependencies, I recommend using the python package manager poetry. You might have to install it before using brew install poetry (on MacOS).

Then, just install the different dependencies:

poetry update

Note: In case you want to use any other methods, a requirements.txt file is also provided.

Quick-start

Once the virtual environment is set up, you can call the script using different parameters.

On MacOS:

poetry run main -bd $(ls ./dataset/*.jpg)

On Windows PowerShell:

python ./main.py $(dir ./dataset/*.jpg)

You can also add options to display:

The input images -dm (display mask)
The boxed differences -bd (display outbound boxes)
The drawed differences -dd (draw differences i.e. the freckles)
The computed ssim model -ds (display ssim)

Notes:

The first given file is the reference which will be use for comparison.
The results are saved under the results/ folder by default.

Example

python ./main.py image-1.jpg image-2.jpg image-3.jpg

image-2.jpg and image-3.jpg will be compared to image-1.jpg.

Back-end logic explained

Let's take the following images:

and:

The first step to do is to compute the ssim difference after having had converted the images from colorized to gray-scaled:

Then, we isolate the differences by binarizing the image:

Then, we remove the noise by applying a mask and removing the outliers i.e. the freckles for which the surface is inferior to an arbitrary number:

Finally, we compute the bouding box for each surface and superpose it to the image:

Troubleshooting

If you encounter a similar error when applying pylint, e.g.:

E1101: Module 'cv2' has no 'imread' member (no-member)

Then, edit the following line in the .pylintrc for:

generated-members=["cv2.*"]

If you encounter the following, related to the use of click:

E1120: No value for argument 'images' in function call (no-value-for-parameter)