I've been looking around for some papers (or info) on this topic.
To avoid a misunderstanding: I'm not talking about finding a supplied pattern in multiple locations.
Repeating patterns can also be understood to mean repeating images. The definition of pattern here isn't abstract. Imagine, for instance, a brick wall. The wall is composed of individual bricks. A picture of the wall is composed of the repeating image of a brick.
The solution must preferably find the largest repeating pattern. Large in this context can be defined two ways: pixel area or number of repetitions.
In the above example, you can cut the bricks in half. In order to make a brick, you can rotate a segment and attach the halves. While the complete brick is the largest repeating image in terms of pixel area, there are 2x more repetitions of half blocks.
Any thoughts?
A number of methods come to mind:
- Fourier Transformation of the image
- Wavelet Analysis
- Autocorrelation
I'd start with fourier analysis: Any shape repeating in a regular pattern in the image creates a very distinct spatial frequency spectrum: One major frequency and some harmonics.
You should probably take a look at this and most of Yanxi Liu's work.
http://vision.cse.psu.edu/research/deformedLattice/Deformed_Lattice_Detection.html
I'm not sure if this is what you're looking for, but I suggest searching for "Texture based segmentation". Take a look at this bibliography, for example: http://www.visionbib.com/bibliography/segment366.html
来源：https://stackoverflow.com/questions/5008040/finding-repeating-patterns-images-in-images