Feature Extraction of stl. files (three dimensional)

Hello,

I have a project in which I'd like to be able to extract features from stl. files of the crowns of teeth (the crown of a teeth is that what you can see when you open your mouth). I have read so many articles in which something similar is being done, but there's never the code for it to be found.

My Input so to say is a stl. file of the surface of the crown, which I want to analyse (feature extraction) to compare it with other teeth (same type of teeth but from another person). The idea is to get the features of every tooth, which should be similar (for example the width of the crown or some characteristic curves) and make a crown with the mean of these features from every tooth.

Can someone give me some advice?

Every bit of help is welcome :D

(I don't want you to do the work for me, just point me in the right direction)

Thank you.

Answer 1

[Gouri Chennuru]

Hi Diego,

You can use generative networks during the process of training data, you can train an autoencoder which takes images from STL file as input, does the process of feature extraction and again convert it into images which is input, you can do it for enough number of images.

Now during the inference, you can take 'n' number of people teeth, extract feature, take the mean and again convert it into images.

Hope this Helps!

[Diego Hens]

Hi Gouri,

Thank you for the answer. I've been looking into this.

My problem with it is the training. I cannot train it to recognize features I myself don't know of. I don't want to tell the program what he should be looking for, but the other way around. I'd love for it to tell me "hey, have you noticed that you can describe almost every tooth with these parameters?"

These sounds to me like a K-means clustering problem. But I'm not sure yet.

The Autoencoder, as far as I have understood it, takes an image (in 2D or 3D, I don't think it matters) breaks it appart in a few steps (those would be the features, right?) and then puts it again together to tell me what kind of an image it is. If it was a number (typical example) it breaks the number appart and in the end it tells me which number it thinks it is.

Ohhh snap, I think I understand now. I can feed him many pictures of many different kinds of teeth (molars, premolars, incisives etc) and train it to recognize what tooth it is. This breaking appart the tooth to recognize which one it is (in the different layers) would be the feature detection, which would give me the values for the features. Have I understood this correctly?

[Gouri Chennuru]

Hi Diego,

Yes, For more information I recommend you to refer autoencoder which performs unsupervised learning of features using autoencoder neural networks.

Regards,

Gouri Chennuru

[Raffaele Felice]

Hi, I've read yours answers and I have a question. In wich way is possible to extract a set of images from an stl?

Thank you, Raffaele.