|
Pink 0.9
|
Calculates the optimal surface segmentation with the maximum flow algorithm [1]. More...
Calculates the optimal surface segmentation with the maximum flow algorithm [1].
Usage: maxflow( (char_image)source and sink, (float_image)constraint image, (int)iterations, (float)tau [, (int)number of threads=0]) -> float_image
Description: In the input, you specify the constraint image 'g' and two subsets of the image 'S'==1 and 'P'==-1. The operator will return a binary flow 'result', which is 0 or 1 almost everywhere. The surface around the volume with value 1 will be the optimum in the sense, that it will contain 'S', not contain 'P' and and it's integral will be the smallest possible.
Generally, we choose the inverse of the gradient for 'g'. Sometimes the inverse is weighted circularly, so gaps in the border are interpolated with circles.
Usually the values, that are known to be in the object (which you want to segment) are set to sink (value 1). You can set sink (value -1) the points, which you know to be outside the object, but generally the border is sufficient.
Types supported: float 2d, float 3d, float 4d, ..., float xd
Category: signal, development ,
References: [1] Appleton, B. and Talbot, H. (2006). Globally minimal surfaces by continuous maximal flows. IEEE Transactions on Pattern Analysis and Machine Intelligence, 28(1):106–118.
1.7.3