matlab - SEM Image Segmentation/Clustering -
i identify clusters of nano structure. here raw image: http://i.imgur.com/pdecw4g.png clusters should this: http://i.imgur.com/zvpaf7p.gif
prefered tool matlab.
background information
overall goal distinguish between foreground , background structures. reconstruct 3d model of porous media, sem images taken slices, binarized solid , pore , placed in row in z-direction. while identifying solid easy, pores unfortunately show solid of subsequent slice. hence, idea verify structures change, i.e. solid of particular slice, , not, being solid of subsequent slices. pixel-wise comparison inaccurate due (nano) drift of structure comparing whole clusters promises better recognition.
suggestions , criticism regarding overall approach welcome!
image segmentation hard problem. different approaches suited different conditions, , i'm not entirely sure "optimum" segmentation you're after. if want separate "reflecting" "not reflecting" in sem, you're right -- you're better off using morphological threshold-based operations you've said in comment above. can use imopen , imclose (in matlab) morphologically open , close image (i.e. connect or shrink structures).
however, in general, no algorithm work across heterogeneous set of images. best bet -- -- use nuclear option , take machine learning approach large amount of training data. looking @ images, isn't clear me ideal solution should -- , that's bad sign.
good luck!
Comments
Post a Comment