# Fourier-Transform for Time Series: About Image Convolution and SciPy | by Yoann Mocquin | Jul, 2023

## Fourier-transform convolution also applies to images

This post is the second of the Fourier-transform for time series, check the first here:

In the first post, I explained how the Fourier-transform can be used to convolve signals very efficiently. I showed that convolution using the Fourier-transform in numpy is many orders of magnitude faster that the standard algebraic approach, and that it corresponds to a certain type of convolution called circular convolution.

In this post, I want to emphasize what the circular convolution means and how it all applies to images. Images are also a good way to extend the 1-dimension intuition into 2 dimensions.

All images were made by the author.

If you’ve ever worked with images for image processing, you most likely have encountered functions to apply convolution. Convoluting images is used everywhere — image enhancement, denoising, segmentation, feature extraction, compression — and is at the base of Convolutionnal Neural Networks, the gold standard of deep learning model to process visual data.

In Python, image convolution can be done quite simply using scipy and its ndimage subpackage. At this point, I recommend taking a quick look at the documentation of the `convolve` function, and then come back here.

The use is very simple: you can pass two images to convolve them together. Let’s see an example: