---

6.3.3 Spatial vs. Frequency domain ¶

With the discussions above it might not be clear when to choose the spatial domain and when to choose the frequency domain. Here we will try to list the benefits of each.

The spatial domain,

• Can correct for the edge effects of convolution, see Edges in the spatial domain.
• Can operate on blank pixels.
• Can be faster than frequency domain when the kernel is small (in terms of the number of pixels on the sides).

The frequency domain,

• Will be much faster when the image and kernel are both large.

As a general rule of thumb, when working on an image of modeled profiles use the frequency domain and when working on an image of real (observed) objects use the spatial domain (corrected for the edges). The reason is that if you apply a frequency domain convolution to a real image, you are going to loose information on the edges and generally you do not want large kernels. But when you have made the profiles in the image yourself, you can just make a larger input image and crop the central parts to completely remove the edge effect, see If convolving afterwards. Also due to oversampling, both the kernels and the images can become very large and the speed boost of frequency domain convolution will significantly improve the processing time, see Oversampling.