segtxt Texture Segmentation using multi-scales multi-channels representation
segtxt [-N N] [-S S] [-W W] [-p p] [-n nr] in mov out
-N N : # images per channel, default 1 (for local scale value)
-S S : standard deviation of the smoothing filter, default 1
-W W : pixel weight for the smoothing filter, default 1
-p p : scalar distance: ABS (p=1) or Quadratic (p=2,default)
-n nr : number of desired regions
in : input Fimage
mov : output Fmovie
out : output segmented Cimage
Cimage segtxt (N , S , W , p , nr , in , mov )
int *N , *W , *nr , *S , *p ;
Fimage in ;
Fmovie mov ;
segtxt is an easy and fast
way to get a segmentation from a textured input fimage.
This module uses directly the modules mschannel and msegct in order to have a direct segmentation of the input fimage. So the user has to enter both the parameters for the mschannel module and the msegct one. The only difference between this module and mschannel is the number of desired regions that the user wants to reach (needed for the msegct module). An output Cimage is then required.
Same as mschannel, the user must enter an output fmovie. This one is used for msegct. Moreover, it can be usefull for the user to see the output fmovie so as to change the parameters if the segmentation is not correctly computed. Just by viewing the output fmovie, the user can fixed the parameters.
In fact, the result of the segmentation depends of the kind of the input fimage. Be not surprised if, at the first time, you don't get what you are expected. Generally, the first parameter N has to be set between 2 and 6. (more if your computer has plenty of memory), it depends if the input fimage contains a lot of litlle details. In this case, N has to be more than 4. Otherwise if your image is not very complex, default parameters could be good.
The others parameters are also important, be carfull in using the parameter S - standard deviation used for the smoothing filter fsmooth, no more than 3, if not, the segmentation won't be good and the computing time huge. Set S to 1 or 2. Same for W (1 or 2). At last p=1 is better if the image is not very contrasted.
For more details, see [KLM94].
Last Modification date : Thu Nov 29 20:23:56 2001