Timeline for Statistical measure for if an image consists of spatially connected separate regions
Current License: CC BY-SA 3.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 14, 2016 at 3:26 | answer | added | GeoMatt22 | timeline score: 3 | |
| Dec 22, 2013 at 6:50 | comment | added | David Marx | You should ask over at the signal processing stack exchange. This is totally their bag. dsp.stackexchange.com | |
| Dec 22, 2013 at 2:02 | answer | added | Pat | timeline score: 0 | |
| Mar 26, 2013 at 4:53 | answer | added | EngrStudent | timeline score: 3 | |
| Jun 9, 2012 at 12:14 | history | edited | Andy | CC BY-SA 3.0 |
added 276 characters in body
|
| Jun 9, 2012 at 8:46 | history | edited | Andy | CC BY-SA 3.0 |
updated based on comments
|
| Jun 8, 2012 at 16:27 | comment | added | Gschneider | What I had in mind was first define some sort of (probably simple) neighborhood structure and compute Moran's I. Then you can compute K, say 200,000, possible permutations of the pixels, computing Moran's I for each permutation. Once you've got these 200,000 Moran's I, see where your observed statistic lies. But, whuber's method sounds easier :). | |
| Jun 8, 2012 at 13:52 | comment | added | Andy | Sorry but I am not sure I follow: are you telling me to permute the image being tested pixel per pixel in some random fashion and then compare the Moran's I value of the permuted image with that of the image being tested? | |
| Jun 8, 2012 at 13:40 | comment | added | whuber♦ | @Gschneider The permutation test is a nice idea, especially because no permutations need to be conducted! It's straightforward to calculate the distribution of the variogram (or Moran's I or Geary's C or whatever) under permutations of the values in the image: there are so many values that the CLT applies. (E.g., the variogram will look like the green points; twice their constant height is the variance of the image values.) The problem becomes more challenging when a "river" pattern needs to be distinguished from other patterns such as a "lake" or "rivers" can have widely varying widths. | |
| Jun 8, 2012 at 13:34 | history | tweeted | twitter.com/#!/StackStats/status/211088744642654209 | ||
| Jun 8, 2012 at 13:15 | comment | added | Gschneider | +1 whuber, Moran's I seems like a good start in this case. Then maybe consider a permutation test to see how "extreme" your image is. | |
| Jun 8, 2012 at 12:59 | comment | added | whuber♦ | Interesting question. One bit of advice that comes to mind immediately is to focus on the short-range part of the variogram: that is the crucial part and will do the best job distinguishing pairs of images like these. (Closely related statistics are Moran's I and Geary's C.) It's hard to give specific advice, though, unless you can more clearly characterize the kinds of images you anticipate processing. | |
| Jun 8, 2012 at 12:22 | history | asked | Andy | CC BY-SA 3.0 |