S
S
Sergey2017-10-25 14:22:47
Image processing
Sergey, 2017-10-25 14:22:47

Create and merge superpixels by color and texture. Can you explain the article?

Hello,
I found an interesting article, which, according to the pictures inside, obtained excellent results in segmenting images based on their color and texture.
The article itself, "Texture superpixels merging by color-texture histograms for color image segmentation", can be downloaded from ResearchGate. What is the meaning of the article:
1. First, using Gabor filters, we obtain information about the texture
2. Based on the information about color and texture, we build superpixels. A new distance metric has been invented that allows you to take into account both color and texture at once
3. Combine superpixels by color (criterion - local histograms normalized by area and texture)
Actually, I didn’t understand the most important things (further notation in the article)
equation 2: is the standard deviation constant for all Gabor filters? Usually they make a different
equation 4: what dimension does Tf get: (M,N, n*k) or (M,N)?
Equation 6: ||(L,a,b)|| - Euclidean norm or something else?
equation 8: Looks like a typo: any non-zero Tf(i) results in zero
equation 9: how to calculate R? Is it just a range of values?
equations 11-13: I didn't understand the idea, where can I read about it?
I would be very grateful for the explanations, if the code works, I will post it on github and indicate the link here

Answer the question

In order to leave comments, you need to log in

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question