# how to detemine this residual matrix ?

10 views (last 30 days)
Commented: Peng Li on 12 Aug 2020
I calculate the cumulative sum of a matrix :Gm,n(i,j)
Now, the next step is '' we adopt the simplest function of a plane-fitting (i.e.Gm,n(i,j)=ai+bj+c ) to fit the trending for each surface Gm,n and determine the residual matrix ym,n(i,j).''
Please, can someone give me the code how to detemine this residual matrix, please ?
##### 2 CommentsShowHide 1 older comment
It is explained like this in the paper.
To estimate the eq 2, I need to determine ''the residual matrix''.
Sincerely, I myself didn't understand, especially the fact that I'm not a mathematician, but I need to program this method. Peng Li on 10 Aug 2020
Interesting. This is an application of the detrended fluctuation analysis (DFA) to a 2D image. Based on what your screenshot shows, it implements the algorithm similarly like being implemented to a time series -- cut into segments based on a time scale s (or here a time-spatial scale), integration (cumulative sum), linear fitting to get residual, and finally there should be a log-log fit between s and F(s).
To answer the specific question you asked, the residual is nothing but the point difference between G_{m,n} and \tilde{G}_{m,n}. This means, the y variable y_{m,n} = G_{m,n} - \tide{G}_{m,n}. F is simply the average of sum of squared y.
Peng Li on 12 Aug 2020
understand and I would just like to encourage you to try it out and ask back here if you get any issues. That way you probably have higher chance to get a solution. I will see what I can do but honestly I have short of time this week and next to meet multiple deadlines, conferences ...