Feature Extraction from an SEM Image

Hi, Okay so here it goes:

I am trying to extract features from an SEM image, now I need to first convert the 2D images into a 3D model. This can be done by using by converting the picture to grayscale format, and then using the grayness values as a heightmap creating a 3D model of the picture, similar to the imageJ picture surface viewer. Now I need to extract features from these models, the problem is that I need to compare two or more images and extract and store these features. In the future I need to use those extracted features in an optimizer to generate a 3D model based on an evaluation criterion. (Ie. first extract the features, then find an optimal landscape with those extracted features as a base using either gp or other metaheuristic optimization schemes). Any input or suggestions are welcome, thanks :D.

Answer 1

[Image Analyst]

I presume you mean multiple 2-D images of the same field of view taken from different angles and then built up a topographic map (iso surface) image rather than a true 3D image like you'd get from CT or MRI where you have internal structures and a value for every single Z value. Anyway, it sounds like you got that part all handled and use surf() do display it. Now you want to extract features but you don't tell us what features. There are lots of features, like Ra, Sa, and numerous other texture features, as well as object-based features like area, perimeter, circularity, equivalent circular diameter, etc.

[Raazi Rizvi]

I have taken a look at the image data I believe the images are not of different angles but of different regions of the sample (see attachments), then again I could be wrong as they were taken by someone other than myself. About the model however I would like to preform supervised feature extraction certain regions of an image to say gain texture related features. The objective is to find optimal features which interact with a water meniscus to give an optimal contact angle. I hope this provides a clearer description of what im trying to do.

[Image Analyst]

I don't see much overlap in those pictures. I could be wrong but I don't know that reflectance of the material at a pixel location can be directly taken as a surface height. If I'm wrong, let me know. I can also ask my SEM microscopists for their opinion on Monday.

Anyway, from your description it sounds like you got the model creation part done. You describe some high level description so I assume you did that and have your model. So you can look up "surface roughness" called SA, and "surface metrology" to get a boatload of other measurements that can be made to describe the surface. If you want object based measurements, like the ones I mentioned, use regionprops. See my Image Segmentation Tutorial in my File Exchange http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862